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.
53 Overview The main focus of this chapter is on activity-based (AB) models. As of 2018, AB models are used by MPOs in 20 of the 25 largest metro- politan areas in the United States, and they are used by MPOs in sev- eral smaller metro areas as well. This discussion assumes some prior knowledge of the concepts and methods used in AB models. Further background can be found in the AB model primer prepared for the Transportation Research Board (TRB) by Chu et al. (2011). The primary difference between AB methods and more traditional trip-based methods is that AB models incorporate a more disaggregated and detailed simulation of travel behavior. The travel of each individual household and person in the region is simulated across the course of a day. Trips are simulated as parts of home-based trip chains (tours), and tours are scheduled within the time available during the day. Travel decisions are simulated as discrete choices based on the model prob- abilities. Using disaggregate discrete choices (rather than multiplying aggregate probabilities, as is done in trip-based models) tends to make the model structure more flexible and able to incorporate several dif- ferent levels and types of choice behavior. As discussed below, this flex- ibility is valuable in incorporating new aspects of travel behavior that may be associated with CAVs. A secondary focus of this chapter is on dynamic traffic assignment (DTA) methods as an alter- native to the more traditional static equilibrium traffic assignment methods introduced in the preceding section on trip-based models. In practice, all AB models are currently applied in com- bination with static equilibrium methods for network traffic assignment. Currently, the main use of dynamic traffic microsimulation is for corridor-scale, project-level analyses that typically employ fixed demand with no feedback of travel times to the travel demand model. Use of DTA methods for region-wide long-range forecasting is still in the initial implementation stages, but combining DTA with AB demand models may become more widespread in the future as the methods and software mature and network data become more plentiful and accessible. Further background can be found in the DTA primer prepared for TRB by Castiglione et al. (2015). For DTA, the trend is toward microscopic dynamic assignment that simulates each vehicleâs trajectory and each driverâs behavior on the network. Rather than use fixed lane capacities and speedâflow relationships, DTA reveals traffic congestion levels and effective capacities through the simulation of how vehicles navigate the roads and intersections. Because no observed data exist on how the introduction of CAVs will affect aggregate speedâflow relationships, the use of C H A P T E R 7 Adapting Disaggregate/Dynamic Models to Address CAVs Chapter Highlights â¢ Provides high-level guidance on accounting for CAVs in AB travel demand models and DTA methods. â¢ Identifies potential model improvements. â¢ Discusses the contexts and approaches for modeling systems: â Sociodemographics, â Land use/built environment, â Auto ownership/mobility models, â Activity generation and scheduling, â Destination/location choice, â Mode choice, â Routing and traffic assignment, â Pricing, and â Truck and commercial vehicles.
54 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles a simulation method that can represent detailed differences in the ways that human drivers and AVs will navigate road networks may be the most promising approach for learning how CAVs will influence traffic capacity and congestion levels. Modeling System Disaggregate modeling systemsâand AB and DTA models in particularâare well suited to evaluating CAVs, with some modifications. The structure of the disaggregate system (as shown in Figure 5) focuses on individual characteristics of the people and vehicles in the system. Much of the recommended guidance requires adding some complexity to the models, so con- sidering the most useful modifications is important. Guidance that will add complexity should only be considered if the features and sensitivities resulting from the model improvement are a high priority for the planning context. Table 7 summarizes model improvements for AB and DTA methods. Sociodemographics Context An attractive feature of AB models is the feasibility of using a large number of household and per- son characteristics in the models. Because each household and person is simulated separately, each can have its own set of sociodemographic characteristics. The distribution of those characteristics Synthetic Population Long-Term Choices Mobility Choices Daily Activity Patterns Tour and Trip Details Trip Assignment Dynamic Traffic Assignment Figure 5. Typical Disaggregate AB and DTA Model Components. Dynamic models are structured to eval- uate individual behavioral responses to changes in the transportation system and thus are well suited to evaluate CAV impacts.
Adapting Disaggregate/Dynamic Models to Address CAVs 55 Model Component Disaggregate AB/DTA Model Improvements Sociodemographics Population synthesizer Control for age and income Population synthesizer Add smartphone ownership and education level Built Environment Urban form Set place type by area type and development type Mobility Vehicle ownership Add CAVs as an option for households to own Vehicle ownership Add purchase cost, incentive policies, parking cost, or accessibility variables to distinguish vehicle type MaaS Add carsharing, ride-hailing, bikesharing memberships Activity Generation and Scheduling Activity generation Lift age restrictions for CAVs, add constraints for persons with disabilities and seniors using conventional vehicles Activity generation Adjust value of travel time (VOT) and review induced demand Activity generation Add representation of empty car trips Destination/Location Choice Work/school locations Integrate with land use model to provide sensitivity Mode Choice Mode choice Add new modes (CAVs, TNCs, shared modes, microtransit) Mode choice Adjust VOT for CAVs Access/egress Add access and egress modes (TNCs, shared modes, microtransit) Mode choice Add dynamic pricing for new modes, adjust parking costs for CAVs Mode choice Adjust age and disability restrictions for CAVs Parking choice Add parking choice model to include off-site parking Routing and Traffic Assignment Dynamic assignment Add vehicle-following and speed characteristics for CAVs Vehicle operations Parameterize vehicle operating characteristics Vehicle operations Track empty vehicles and their travel characteristics Dynamic assignment Simulate different levels of CVs in mixed traffic Dynamic assignment Simulate nonrecurring congestion with/without CAVs Pricing Cost models Determine cost per mile for each new mode by time period Parking costs Adjust parking cost as demand shifts away from high-cost areas Truck and Commercial Vehicles Supply chain Adjust cost and time for CAVs Truck touring Adjust driver hours of service for CAVs Truck touring Add pick-up and delivery services by TNC Table 7. Summary of model improvements for AB and DTA models. across the population is controlled through population synthesis. In the synthesis process, distri- butions of key variables such as income, household size, number of workers, and age group are controlled at a detailed geographic level such as Census tract, Census block group, or TAZ. Approach Age group is currently an important determinant of the likelihood of using new transporta- tion options such as Uber, Lyft, car2go, and other TNCs. Younger households and persons are more likely to use these new options, particularly for utilitarian purposes. After age is taken into account, people with higher incomes are also more likely to use TNCs. Age and income will
56 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles also be important variables for modeling CAV use. As discussed in more detail below, age- and income-related preferences may be quite different for owning a private CAV as opposed to using CAV-based TNCs. As a result, it is useful to set age and income as controlled variables to ensure that the synthesized population is more accurate. An advantage of population synthesis is that additional characteristics of the population can be added when data become available to control for these characteristics. One example is the strong relationship between smartphone ownership and other new technologies, such as AVs. The population synthesizer could be adapted to identify smartphone ownership for households that will have access to MaaS. Another personal characteristic that influences smartphone own- ership (and possibly adoption of new technologies) is education level, so this could also be added to the population synthesizer. Land Use/Built Environment Context AB models use a wide variety of built environment variables as inputs. Some of these, such as intersection density and mixed-use indices, are important in modeling the use of walk and bike modes and trip-chaining behavior. Destination choice and location choice models use several categories of employment and student enrollment as attraction variables, as well as other vari- ables such as parks and open space and single- and multifamily housing units. Place type is another means of identifying the characteristics of the built environment for use in travel models. Place types can be set as a combination of area type (central business district, urban, suburban, rural) and development type (residential, employment, mixed use, transit oriented, greenfield) (Outwater et al. 2014). Place types are an important indicator for mobility services, which tend to be deployed in business districts and urban areas. Approach If the AB travel model is integrated with a land use model, the same type of accessibility measures should be fed back to the land use model, typically in the form of mode/destination logsums across all available modes and destinations from a given residential location. Depending on the scenario, these factors could act in opposite directions or in the same direc- tion. Thus, the net effects on accessibility, commuting distances, and residential patterns are not obvious, but it is important to include as many of these factors as possible to avoid obtaining a one-sided result. Auto Ownership/Mobility Models Context Like many advanced trip-based models, AB models predict auto ownership as a longer-term decision that conditions day-level travel choices. Some AB models also include simple models of other possible longer-term mobility factors, such as transit pass ownership, availability of free or subsidized parking at the workplace, toll transponder ownership, or bicycle ownership. Approach To represent private CAV ownership, the auto ownership model can be enhanced to predict both the number and type of vehiclesâCAV or conventional. The simplest approach is to make
Adapting Disaggregate/Dynamic Models to Address CAVs 57 it an all-or-nothing choice between owning CAVs or conventional vehicles. The left side of Fig- ure 6 shows this structure in which a household first chooses what type of vehicles to buy and then how many. A more complex approach would be to also allow mixed ownership, as dem- onstrated in the right side of Figure 6, in which a household first chooses how many vehicles to purchase and then what type. Most current AB models used in practice enforce travel schedule consistency for persons but not for vehicles. As a result, these new data on vehicle type could be used initially to influence travel behavior rather than constrain individual travelers to specific vehicles. The types of variables that can be used to model the choice between owning CAVs and con- ventional vehicles include â¢ Sociodemographics such as age and income, as mentioned above; â¢ Relative purchase costs of CAVs and conventional vehicles; â¢ Incentive policies to encourage CAV ownership and conventional vehicle trade-in; â¢ Relative operating costs of CAVs and conventional vehicles, if any substantial differences are assumed; â¢ Accessibility/utility difference of using a CAV for homeâwork commutes and perhaps other types of destinations (e.g., a reduced disutility of in-vehicle time for CAVs would make pur- chase more likely for those with long commute times); and â¢ Parking costs paid at the residence or workplace(s) that might be avoided with a self-parking CAV. Transportation planners might also consider including an interaction between the number and type of vehicles owned. For example, planners could assume that households choosing to purchase a CAV would be more likely to own one vehicle rather than two vehicles, because the CAV could serve as a private taxi to take both would-be drivers to their destinations. This assumption, along with most of the assumptions used to represent CAVs in the models, is clearly speculation and not based on any data from actual purchase decisions. Thus, it is important to use the models initially in an exploratory way to investigate the implications of different Auto Ownership Conventional None 1 vehicle 2 vehicles 3+ vehicles CAV None 1 vehicle 2 vehicles 3+ vehicles Auto Ownership None 1 vehicle Conventional CAV 2 vehicles Conventional CAV Both 3+ vehicles Conventional CAV Both Figure 6. Potential nesting structures for auto ownership models with CAVs.
58 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles assumptions. It is also important to refine the assumptions over time as new data become avail- able from stated preference surveys and (eventually) actual purchase decisions. For the other types of mobility models, planners could consider modeling membership in a CAV sharing club or group. The high purchase cost and taxi-like usage characteristics of CAVs may encourage shared ownership across households that uses some type of shared scheduling app similar to TNCs but is limited to a small set of owners. This would presumably offer a shared purchase and maintenance cost, but with the possible inconvenience of not having the vehicle available when needed and having to use a TNC or other mode instead. Once the number of possible users of a shared CAV service becomes large enough (i.e., approaching todayâs Zipcar and car2go systems), the differences between CAVs and other TNCs such as Uber and Lyft will largely disappear. In either case, the vehicle will come and pick up the user, and it will be possible to schedule use in advance. The main remaining difference may be in the cost structure, with some operators requiring a membership price in return for lower per-trip cost (and perhaps greater availability). If the only difference is in price structure, it is not obvious that it will be worth the effort to model membership-based CAV sharing and Uber- and Lyft-type CAV sharing as separate options. Instead, they could be modeled as a single mode with an average price. Activity Generation and Scheduling Context Current activity scheduling models predict activities and trips for each person in a household, subject to time constraints and vehicles available. Some AB models also jointly schedule travel across household members constrained by the same time and vehicle availability. Vehicles are not typically tracked in AB models, so there is no constraint on the use of a specific vehicle, just constraints on the total vehicles available for the household. AB models allow household members to choose between multistop tours to complete sev- eral activities and several individual tours to complete the same activities. These trade-offs will become more complex as travelers choose whether to continue to trip-chain or send the vehicle to complete some activities. Activity-scheduling models are sensitive to accessibility, so more congested time periods and higher tolls or parking would tend to push peak period travel into off-peak periods. Improved accessibilities and reduced travel cost will also induce overall demand for travel, something that AB models measure directly. Approach Some aspects of current activity-generation and -scheduling models may need to be recon- sidered in light of the increased availability of travel without a driver. Age restrictions or refer- ences to persons with driver licenses will only be relevant for conventional vehicles. Lifting these constraints will induce travel from these populations. Some AB models may not currently limit driving for these populations but are calibrated to data that do. Adding these constraints to the models and then recalibrating could provide a means to support the estimation of induced demand. Another aspect of autonomous and connected travel is that travel is less onerous overall. This stems from travelers being able to multitask, vehicles traveling closer together, and the ability for travel in a vehicle to begin and end at places other than where the vehicle is parked. Reductions in travel time and cost will tend to increase the amount of travel that people engage in, given latent demand for travel. This type of induced demand is represented by the accessibility measures included in activity generation models.
Adapting Disaggregate/Dynamic Models to Address CAVs 59 The current time-scheduling component focuses on schedule constraints for persons rather than vehicles. Adding new components that track vehicles would allow the models to constrain travel on the basis of which household vehicle is availableâa CAV or a conventional vehicle. This change would add considerable complexity to the modeling system but would provide a more direct estimation of ZOVs. This is likely an AB model improvement that would occur in the future, rather than in initial versions of CAV modeling. AVs offer opportunities for household members to optimize travel more directly, as vehicles can reposition themselves to serve multiple household members. This practice may reduce trips if household members carpool to multiple destinations. This practice may also increase trips if household members return the vehicle home after each trip. Activity-generation models will need to recognize that some movements will be this actual repositioning and keep track of these zero-occupant movements to assess impacts on the transportation system. Destination/Location Choice Context Destination choice is deployed in AB models in two ways: first, to identify usual work and school locations, and second, to identify destinations for each activity for each household mem- ber over the course of the day. Most work and school trips go to the usual location each day, but sometimes travelers attend a business meeting or work at a different location. AB models generally treat residential location as fixed on the basis of population synthesis in the base year, and perhaps modified by a land use model for future years. In contrast to trip- based models, AB models predict the usual work location for each worker and the usual school location for each student as longer-term decisions, and then predict the travel behavior for a specific day conditional on those usual work and school locations. Destination choice models rely primarily on level-of-service characteristics (e.g., time, cost, distance) to identify preferred destinations, but these models are also sensitive to demographic and built environment factors. Travel time and cost variables are typically represented as mode choice logsums, so changes in accessibility for any mode will affect the attractiveness of a destination. Approach Although CAVs are expected to have several impacts on destination choice, no adjustments have been made to the current implementation of destination choice models. The expected impacts on destination choice models include the following: â¢ In most places in the United States where the auto is already the dominant mode, the main effect of reduced in-vehicle VOT for CAVs will be longer trips, nonmandatory trips in the short term, and commute trips in the longer term. This is the primary way that VOT affects VMT in the models. â¢ CAVs could make it more attractive for people to choose a job farther from their home or choose a residence farther from their workplace or both. Some other effects, however, might work in the opposite direction. The choice of usual work location is part of the AB model, and it is important that the model include accessibility measures such as mode choice logsums (the expected utility across all available modes) between home and alternative work locations. â¢ If the dominant mode of CAV use is TNC-based, then a higher price per mile than for private AVs may mitigate longer trip lengths. If customers begin to trade off the purchase of an auto- mobile with the prospect of using on-demand mobility services, then the price per mile for the private AV will be higher than that for the on-demand service.
60 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles â¢ CAVs will increase the convenience and attractiveness of destinations in parking-constrained areas. Parking cost is currently an impediment to driving to these areas, and both AVs and TNCs will avoid parking charges. Some AB models currently assign joint travel among household members, and this behavior is likely to change as AVs offer opportunities to travel between destinations to serve additional household members. Current restrictions on joint travel and mode switching should be recon- sidered in light of the flexibility that CAVs offer. Mode Choice Context Current AB models contain a tour-based and a trip-based mode choice. New modes for on- demand servicesâTNCs (solo), TNCs (shared), carshare, bikeshare, microtransitâare not well represented, although a few AB models are adding TNCs as a new mode. The current model structure and mathematical formulations are sufficient to consider these new modes. Current AB models do not distinguish vehicle types (e.g., hybrid, electric, autonomous, con- nected). Vehicle choice models would be needed to assign a specific vehicle to a specific person in the simulation, but vehicle type could also be added as a nest within the mode choice model. Current mode choice models may restrict driving to persons over 16 years of age, but the major- ity do not segment the population by persons with disabilities or seniors who cannot drive. These populations will have new access to auto modes that are not available today. Approach New mode choice models will need to incorporate relevant new modes such as TNCs, car- share, bikeshare, and microtransit. Auto modes can expand to include taxi, TNC, carshare, and CAV. Transit modes can expand to include on-demand microtransit and taxi, TNC, carshare, and bikeshare as new access and egress modes. The bike mode can expand to include bikeshare. Most of these options will require new assumptions about the use of these new modes; for example, people can pick up a car in the middle of a tour to complete one or more trips. At this point, it is unclear how to represent these new modes in a nested logit model structure, as the data with which to estimate nested mode choice models are limited for these new modes. Thus, care should be taken to test the impacts of different nesting structures. Not having to drive the vehicle may make in-vehicle time in a CAV less onerous than time spent in a conventional vehicle and thus reduce the value (disutility) of travel time (i.e., VOT). Values of time spent waiting, walking, driving, or riding are all separately evaluated to ensure that the model simulates behavioral choices correctly. The expected reduction in VOT for in-vehicle time spent in a CAV can significantly affect the choice of CAV over another mode. Current mode choice models represent access modes as walk and drive (and sometimes tran- sit), and egress modes are typically limited to walking. Some models represent park-and-ride and kiss-and-ride as separate drive access modes. New modes will offer multiple new access and egress modes, such as TNCs and bikeshare. Expanding the egress mode options beyond walking and transit may improve the convenience of transit. Dynamic pricing of new modes should be represented in AB models, which currently rely on 5â10 different time periods for estimation of travel times and costs. As a result, calculating Today, it is unclear how to represent new modes such as TNC, carshare, bikeshare, and microtransit in a nested logit model structure, as limited data are available for these modes.
Adapting Disaggregate/Dynamic Models to Address CAVs 61 aggregate prices by time periods will be more effective than trying to expand the number of time periods, owing primarily to the added complexity and processing time required. Current pricing models for TNCs offer a carpool mode at a lower price; this is another complexity that may be difficult to implement as a separate mode. The trade-offs in pricing for buying a new vehicle compared with using an on-demand service may also change over time as customers begin to understand the per-mile costs of owning versus subscribing to a service. The per-mile travel cost for privately owned CAVs may differ substantially from the per-mile cost of CAV-based TNCs. The introduction of CAVs will affect all the auto modes (conventional vehicle and CAV for privately owned and on-demand services) in different ways. The use of privately owned CAVs will be quite different from the use of conventional vehicles, since the vehicles can be reposi- tioned for multiple travelers. In addition, travelers can engage in multiple activities in the vehicle because there is no need to drive. Travelers can also expect no parking cost and higher conve- nience in congested areas, but they may see a higher operating cost if the vehicle is sent home or to pick up another traveler. Moreover, planners must remember that on-demand services will be available in some areas but not others. The added complexity of identifying each new mode in the mode choice model will depend on the importance of each new mode to the region of interest. New modes offer tremendous choice for mobility-challenged populations (children, seniors, and persons with disabilities), because owning a private CAV can serve their travel needs better than conventional vehicles that require a driver. Many of the new AB models incorporate age as an input, so the models can be adjusted to allow younger and older travelers to travel in a private CAV or on-demand CAV. Since current models tend not to identify the population of persons with disabilities, this change could be made to the underlying demographics so that these travel- ers could have more options in mode choice. Again, the complexities of adding these constraints should be weighed against the additional detail provided. Routing and Traffic Assignment Context Several agencies have been developing DTA methods applied at a regional or corridor scale to improve the commonly deployed static traffic assignments. DTA methods are disaggregate, simulating each vehicleâs movement on the road network. Integrating DTA methods with AB models to complete the disaggregate modeling system has been researched for many years. Currently, several test beds are under development but have not yet been used in planning applications. DTA is advantageous for evaluating CAVs because the operational characteristics of CAVs are different from those of conventional vehi- cles. The differences can be simulated to better understand the opera- tional characteristics of mixed-flow and CAV-only operations. DTA simulations assign vehicles by user class, which can be defined by type of vehicle (e.g., car, truck, CAV), number of occupants, or type of trip (e.g., commute versus discretionary travel). The number of user classes adds complexity and processing time and should be selected to identify the most important characteristics of the infrastructure and policy improvements under consideration. At least two classes are important to understanding new technologies: con- ventional vehicles and CAVs. If zero-occupant CAVs need to be tracked separately, then another user class for these vehicles can be included. DTA methods are disaggregate, simu- lating each vehicleâs movement on the road network. DTA is advantageous for evaluating CAVs because simula- tion can be used to better under- stand mixed-flow versus CAV-only operations.
62 Updating Regional Transportation Planning and Modeling Tools to Address Impacts of Connected and Automated Vehicles Approach DTA can identify different vehicle-following distance and speed characteristics, depending on the level of automation. If these vehicles are traveling in mixed traffic, then overall travel times (i.e., skims) are generated for all vehicles. If separate facilities are considered for CAVs, then it is important to separately track the travel times for CAVs and conventional vehicles. Given the complexity of DTA models, especially DTA models that are integrated with AB models, there is a benefit to applying the DTA model to simulate operating characteristics for use in regional travel demand models in place of an integrated ABâDTA model. The operating characteristics can then be regressed to estimate functions that do not require simulations (e.g., simulating wait times for a TNC trip as a function of land use density or simulating the freeway speed of a CAV on a con- gested facility as a function of the speed of a conventional vehicle in a separate lane). This process can be effective in specifying changes in capacity that result from CAV operations and ultimately for redefining volume-delay functions for CAVs that can then be applied in static assignment. Another area in which DTA models can provide insight is the repositioning of CAVs when not in use. This is also referred to as ZOVs when traveling without a passenger. This repositioning will occur for both privately owned and fleet-operated CAVs. Repositioning to pick up another passenger can be directly simulated from the drop-off and pick-up locations of each passenger. Repositioning that is used to avoid parking costs can be more difficult, as the parking locations for these trips may not be predetermined. Nonetheless, the inclusion of these ZOV trips is criti- cal to understanding congestion in the system. These DTA simulations can offer insight into the performance impact on the transportation system and, in turn, the impact on traveler behavior. Connected infrastructure, such as V2V or V2I, provides a significantly different set of operat- ing characteristics than conventional vehicles as the percentage of connectivity increases. These operational characteristics can be better understood by simulating different levels of CAV adop- tion for vehicles and infrastructure. Optimized signal timings or route switching can be simu- lated to produce functions that describe these operations. Another area in which DTA models provide additional detail is operations during an accident or other nonrecurring event. Delays associated with nonrecurring events are significant, and CAVs are expected to significantly reduce accidents once human drivers are no longer part of the traffic flow. DTA models can produce the operational improvements that lead to time and cost savings for all travelers as a result of fewer accidents. Pricing Context Pricing has been identified as an important policy approach to encouraging shared AV use (Zmud et al. 2017). Current AB models incorporate pricing as an input for determining whether to travel, where to travel, and how to travel (which mode). These models are iterated to address congestion effects but are typically too time consuming to iterate for dynamic pricing effects. Current vehicle ownership models do not incorporate pricing as an input since the vehicle type is not critical to the performance outcomes. When households are faced with new mobility choices, it may become more important to recognize purchase cost for vehicles as a trade-off for the cost of using a mobility service. Approach As new data on TNC operations and costs become available and new research on traveler choice for these new services is conducted, planners can develop pricing inputs for new modes.
Adapting Disaggregate/Dynamic Models to Address CAVs 63 These pricing inputs can approximate dynamic pricing by time period. Since costs for new services are changing rapidly and the cost for CAVs is not yet set, planners should test many dif- ferent pricing options to evaluate the range of expected outcomes. Modeling of current vehicle purchasing decisions should incorporate cost and demographics to directly represent the impor- tance of cost in determining whether to purchase a CAV or a conventional vehicle and the trade- off with mobility services as a subscription. Truck and Commercial Vehicles Context Disaggregate microsimulation models for goods movement and commercial vehicles repre- sent the supply chain for products from producer to consumer and pick-up and delivery services to deliver products to their final destination. These freight forecasting models follow a series of sequential steps to simulate commercial vehicle movement: â¢ Supply chain: â Firm synthesis, â Procurement markets, â Distribution channels, â Shipment size and frequency, and â Modes and transfers. â¢ Truck touring: â Vehicle and tour pattern choice, â Number of tours and stops, â Stop sequence and duration, â Delivery time of day, and â Truck assignment. Most current models focus on truck assignments. Rail simulation models address operations for carloads, but they are not typically applied in a planning context. Given the continued driver shortage in the trucking industry and the potential to lift restric- tions on driver hours of service, CAVs could dramatically affect the shipment of goods by truck. Current disaggregate models are well positioned to adapt to these changes. Approach Current supply chain models do not require any structural changes to represent CAVs. The primary impact will be to adjust the time and cost for trucks, which could in turn increase mar- ket share for trucks. Early pilot tests have shown significant cost savings for connected trucks, which can be incorporated as a new input. Current truck touring models restrict drivers to regulated hours of service, and if regulations change with CAV technology, then the models can adapt to recognize this flexibility. If current models do not recognize this restriction, they should first be adapted to restrict drivers, and then the restriction can be lifted to represent the new rules. Current truck touring models identify pick-up and delivery services operated by shippers or carriers rather than TNC-style delivery services using noncommercial service vehicles. The use of CAVs and/or drones for pick-up and delivery could shift the vehicle and tour pattern choice models, along with the number of tour and stop models. Truck touring models should be adapted to incorporate this additional source of pick-up and delivery services.