Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles (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.

71 Using modeling and simulation, the research team evaluated the specific conditions and factors that influence the performance of CAV applications when such users are in a DL. The team used a four-step evaluation approach to evaluate the performance of CAV applications under differ- ent conditions and to quantify factors that influence these impacts (see Figure 5.1). This chapter expands on each of the four steps of the evaluation process. As was discussed in Chapter 1, the team conducted extensive research to identify the scope of this evaluation, includ- ing the test sites, the CAV applications to be included, and the specific performance measures and research questions. 5.1 Develop Baseline Models The team ranked the various simulation testbeds that were available based on evaluation criteria pertaining to (1) case study site characteristics, (2) managed lane characteristics, and (3) CAV modeling feasibility. In the end, two case study sites were selected: the I-66 corridor in Northern Virginia and the US-101 corridor in San Mateo County, California. The next step was to develop baseline simulation models that could replicate the real-life traffic behavior of the two selected sites. The models for both simulations were developed using PTV VISSIM micro- simulation software. The first step to effective analysis using modeling and simulation is to develop baseline models that are representative of the field conditions. As per the FHWA’s Traffic Analysis Toolbox, the microsimulation model development process consists of several steps, as shown in Figure 5.2. The four steps highlighted in grey in the figure are required to develop a well-defined baseline- calibrated model that provides a good representation of the field conditions. As shown in the figure, once the pre-calibrated model (which includes data about the lane geometry, turn permissions, infrastructure, and traffic control features as well as traffic demand and vehicle types) has been developed, the model is iteratively adjusted such that the mea- sured performance resembles the field conditions. This includes comparing vehicle volumes and speeds at differing locations of the network in order to match the simulation with the field conditions and to ascertain whether congestion occurs at the right places along the network. The models adopted by the project team for this evaluation—I-66, in Northern Virginia, and US-101, in San Mateo County, California—were at differing levels of maturity. For example, the I-66 model required complete calibration based on field data to conduct the analysis, whereas the US-101 model had already been calibrated during previous projects. This chapter provides details on the calibration process and calibration results for the I-66 corridor, and provides the baseline system performance for both testbeds. These performance measures formed the base- line to which application performance was compared, as described in Chapter 6. C H A P T E R 5 Analysis and Evaluation Approach

72 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles 5.1.1 I-66 in Northern Virginia Interstate Highway 66 (I-66) in Northern Virginia was the primary case study site for this project. The I-66 corridor is an east-west corridor taking traffic between the suburbs of Northern Virginia and Washington, D.C. The selected test site represents the segment of I-66 between the I-495 interchange and the SR-234 interchange, spanning approximately 13 miles, as shown in Figure 5.3. The corridor has an AADT of 150,000 to 160,000 vehicles and includes one peak-hour peak-direction HOV-2+ lane in addition to two to three GPLs and a peak-hour peak-direction shoulder lane. The freeway corridor has a posted speed limit of 55 mph. This suburban test site includes six interchanges and two dedicated on-and-off ramps for the HOV lane, separate from the GPLs. The average distance between interchanges is approximately 1.2 miles, yielding 0.6 miles and 2 miles of minimum and maximum interchange spacing, respectively. The test site experiences recurring congestion caused by high directional daily demand every weekday for the eastbound lanes (i.e., traffic moving toward Washington, D.C.) during the a.m. peak and the westbound lanes (i.e., traffic moving toward Fairfax, Virginia) during the p.m. peak. Between 2:00 p.m. and 8:00 p.m., the traffic volumes of the testbed range from 900 vphpl to 2,100 vphpl, including approximately 1,500 vphpl of peak HOV traffic volumes (Lu et al. 2014). The I-66 managed lanes are single-lane, time-of-day, HOV-2 lanes in both eastbound and westbound directions and are located in the left-most lanes. User-type restrictions along the existing HOV-2 allow for all vehicle classes with occupancy requirements of two or more occu- pants per vehicle. The existing managed HOV-2 lanes operate on a time-of-day basis with restrictions applying during a.m. and p.m. peak periods on weekdays. No physical barrier sepa- rates the managed HOV-2 lanes from the mixed use lanes. Currently, only double solid white lane markings are used for lane separation and to indicate no lane changing/no access. Access points between the dedicated HOV-2 lanes and the mixed use lanes are permissible only along sections with dashed lane striping. The I-66 also has hard-shoulder running lanes on the right-most lanes in both directions. The hard-shoulder running lanes operate from 5:30 a.m. to 11:00 a.m. in the eastbound direction and from 2:00 p.m. to 8:00 p.m. in the westbound direction. Lane utilization is indicated via VMS that show a green arrow for permitted use and a red cross for closed unless exiting. The I-66 Northern Virginia VISSIM base model was obtained through the U.S.DOT OSADP (U.S.DOT n.d.b). However, the base model calibration was not suitable for this project. Figure 5.1. Overall evaluation process. Figure 5.2. Microsimulation model development and application process.

Analysis and Evaluation Approach 73 Therefore, the project team collected field traffic data for vehicle counts, incident records, spot speeds, and probe data from the Virginia Department of Transportation (Virginia DOT). Figure 5.4 shows the data used for calibrating the network and represents a typical day peak- hour traffic at 1-hour resolution for the HOV and non-HOV lanes for October 2016. The simulation model was calibrated to field observations using the traffic analysis toolbox performance measure for model validation developed by Dowling et al. (2004) and the GEH statistic, an empirical formula devised by Geoffrey E. Havers. The GEH statistic compares the simulated and real-world hourly traffic volumes. Figure 5.5 shows a comparison of the field data with the model travel-time data. In Figure 5.5, the red line represents the field travel time data and the blue region represents the acceptable ±10% margin of error. The field travel time was computed from the probe vehicle speed data provided by the Virginia DOT from the RITIS database. These data were computed on an hourly basis; therefore, the red line is plotted from five data points representing 3:00 p.m., 4:00 p.m., 5:00 p.m., 6:00 p.m., and 7:00 p.m. The VISSIM model was simulated and recorded over multiple speeds to achieve statistical significance, and the travel time was averaged and represented by the blue line. The green region represents the boundary of all the travel time for individual random speeds. The calibration results shown in Figure 5.5 indicate that all model travel times are within the acceptable margin of error and the average travel time replicates field travel time by increasing from the start until it peaks (at approximately 5:00 p.m.) and subsequently declining until the end of the simulation time frame. Table 5.1 shows the calibration results using traffic counts as a performance measure with the GEH statistic to validate the model’s ability to replicate existing traffic patterns. The results shown in Table 5.1 indicate that, out of 32 comparisons, only two yield a GEH value larger than 5.0. Taken together, the cases result in 93.75% target values falling within the acceptable 85% of freeway mainline links and within the maximum GEH value of 5. Havers’ Source: NCHRP 20-102(08) project team; base map from www.openstreetmap.org. Figure 5.3. I-66 case study site.

Source: NCHRP 20-102(08) project team; base map from www.openstreetmap.org. Figure 5.4. Traffic volumes and speed for calibrating the I-66 model.

Analysis and Evaluation Approach 75 Figure 5.5. I-66 westbound travel time field and simulation data with calibration threshold. Field Vehicle Counts Simulation Vehicle Counts GEH Statistic Location 3:00 p.m. 4:00 p.m. 5:00 p.m. 6:00 p.m. 3:00 p.m. 4:00 p.m. 5:00 p.m. 6:00 p.m. 3:00 p.m. 4:00 p.m. 5:00 p.m. 6:00 p.m. W es tb ou nd East of VA-243 4,194 3,982 3,810 3,723 4,313 3,772 3,855 4,023 1.8 3.4 0.7 4.8 VA-234 and VA-123 3,432 3,141 2,926 2,947 3,451 3,210 2,759 3,059 0.3 1.2 3.1 2.0 VA-123 and US-50 4,161 4,154 3,890 3,926 4,122 4,085 4,171 4,242 0.6 1.1 4.4 4.9 VA-286 and VA-28 4,670 4,056 3,738 3,983 5,047 3,934 4,041 3,971 5.4 1.9 4.9 0.2 Ea st bo un d VA-286 and VA-28 4,484 4,415 4,509 4,312 4,387 4,500 4,224 4,320 1.5 1.3 4.3 0.1 VA-123 and US-50 4,894 5,244 5,424 4,959 5,059 4,932 5,523 4,909 2.3 4.4 1.3 0.7 VA-234 and VA-123 2,784 2,955 3,102 2,792 2,589 2,942 3,235 2,656 3.8 0.2 2.4 2.6 East of VA-243 4,518 4,593 4,754 4,248 4,772 4,189 4,688 4,408 3.7 6.1 1.0 2.4 Table 5.1. I-66 field and simulation vehicle counts with the GEH statistic.

76 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles GEH statistic functions as an empirical parameter similar to the chi-squared statistic that compares two sets of traffic volumes. 5.1.1.1 Baseline System Performance This section expands on the baseline system performance. Although the model was cali- brated on both eastbound and westbound directions, the research team’s performance mea- surement focused on the westbound direction because that was the direction of peak traffic flow. Figure 5.6 shows the calibrated baseline freeway speeds of I-66 westbound with data col- lected using data collection points spaced at 0.1-mile increments. The speed profile shows sev- eral bottleneck locations near the interchange locations, with traffic congestion also forming from the network downstream location. The spatio-temporal speed profile is a useful technique to demonstrate how the DSH application can help in reducing congestion and shockwaves over the network. Figure 5.7 shows the baseline average hourly throughputs of the freeway. The throughput shown in this figure is an average of 12 cordon lines representing different sections of the freeway. The traffic changes over both spatial and temporal dimensions, but for easier demonstration, only the temporal changes are shown. Spatial changes were averaged into Figure 5.6. Speed profile, I-66 westbound. Figure 5.7. Baseline freeway throughput for westbound I-66.

Analysis and Evaluation Approach 77 computation of the throughput values. The time-series in Figure 5.7 shows that the volumes increase between 5:00 p.m. and 6:00 p.m. and then steadily decrease, as was expected from the field data. Given that the CACC application aims to improve the capacity of freeway lanes, average freeway throughput was considered an effective way of demonstrating the impact of CACC-equipped vehicles in the network. Figure 5.8 shows the baseline performance measures for the entire I-66 network in terms of average vehicle delay, average vehicle speed, total vehicle travel time, and total system throughput. 5.1.2 US-101 in San Mateo County, California Although the I-66 corridor provided enough precedence to test various scenarios in a CAV DL setting, the project team also used a secondary case study site to supplement some of the analysis that was performed. Specifically, having a second case study site enabled understanding of any measured variability that could be attributed to differences in operational characteristics, geo- metric/geographic characteristics, driving behavior, demand, and so forth. The secondary case study site chosen by the study team represents the US-101 corridor located in the County of San Mateo, California, and stretches from Redwood City to the City of Burlingame. The length of the modeled US-101 freeway facility is approximately 8.5 miles, with a parallel arterial, El Camino Real (SR-82), of similar length. The model allows diversions for vehicles from the US-101 free- way to the SR-82 via seven interchanges. The extent and coverage of the US-101 corridor model is illustrated in Figure 5.9 (Booz Allen Hamilton 2016). Figure 5.8. Baseline performance measures for I-66 network.

78 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles The US-101 freeway carries between 200,000 and 250,000 AADT, of which 15% to 25% con- sists of HOV-2+ vehicles. El Camino Real carries between 25,000 and 50,000 average daily traffic. The proposed HOV-2+ DL facility will have continuous access throughout the US-101 freeway. Currently, the HOV lanes are located at the southern portion of the freeway and end at the interchange with Whipple Avenue. The freeway has a posted speed limit of 65 mph. The model was previously utilized by the U.S.DOT to conduct an impact assessment of the DSH application using connected vehicle modeling. As the secondary site, the US-101 case study site underwent minimum modifications for use in this project. Modifications were made to the classification of vehicle classes and coding of data collection measurements to provide the necessary inputs to the CAV applications. No modifications were made to the model in relation to the portion of the El Camino Real arterial to accommodate the DL on US-101. 5.1.2.1 Baseline System Performance This section describes the baseline performance of the modeled US-101 corridor. For full details on the data collection and calibration process with respect to this case study site, readers are encouraged to refer to the calibration report for the San Mateo Testbed by Yelchuru et al. (2016). Figure 5.10 and Figure 5.11 show some of the observed performance measures for the base model averaged using five runs (to create statistically significant results). Specifically, Figure 5.10 shows the baseline corridor travel time as a function of p.m. peak intervals for the US-101 northbound direction. Other system-wide performance measures, such as average vehicle delay, average vehicle speed, total vehicle travel time, and total system throughput, are shown in Figure 5.11. Source: NCHRP 20-102(08) project team; base map data © Google. Figure 5.9. US-101 case study site coverage.

Analysis and Evaluation Approach 79 Figure 5.10. Baseline performance in northbound (peak direction) travel time. Figure 5.11. Baseline performance measures for US-101.

80 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles The analysis period used was between 2:30 p.m. and 7:30 p.m. It can be seen from the figures that the peak congestion occurs between 4:00 p.m. and 6:00 p.m., as represented by the reduc- tion in vehicle speeds or the increase in vehicle delays. 5.2 Integrate CAV Application Models The project team reviewed several longitudinal and lateral driving behavior models that could be considered in a connected and automated environment to evaluate the feasibility of priori- tized and exclusive DL usage for CAVs. For this evaluation, priority was given to freeway-based applications, given that dedicating lanes on arterials presents significant challenges in quantifying the impacts due to variability in arterial design, signalized and non-signalized operations, lack of mature CAV applications for arterials, complex V2V and vehicle-to-pedestrian interactions, and so forth. Of the freeway applications evaluated, the two highly researched applications that could potentially be deployed in a DL setting were CACC and DSH. PTV VISSIM provides two ways to code in customized vehicle controls into simulation: • Component Object Model (COM). The COM interface defines a hierarchical model in which functions and parameters of the simulator originally provided by the default models can be manipulated by programming; and • External Driver Model (EDM). This Dynamic Linked Libraries (DLLs) interface can be used to define driver models such as vehicle behavior, acceleration, position within a lane, desired speed, look ahead distance, and so forth. For this step, the research team integrated these two methods so that they could be used in conjunction to enable modeling of CAV applications. The balance of this section describes the CAV applications and the modeled features. 5.2.1 CACC CACC has been widely researched, and several algorithms exist to provide control logic for equipped vehicles to platoon at short headways. The project team selected the VISSIM-based CACC API developed by Lee et al. (2016a) as a starting point to model the CACC application for this project because of its maturity and adaptability to the project needs. Lee’s model, originally developed in 2014, was enhanced in 2016. Based on work by TNO (The Netherlands Organization for Applied Scientific Research), the model utilizes an Enhanced Intelligent Driver Model (E-IDM) to simulate CACC string behavior and char- acteristics (Schakel et al. 2010). For this study, Lee’s model was applied to three types of car-following behavior: (1) VISSIM’s default car-following model (i.e., psychophysical car- following) for non-CACC drivers; (2) the IDM for the ACC driver to represent the Leader vehicle of a CACC platoon; and (3) a customized IDM to deal with CACC longitudinal maneuvering for all the follower vehicles. Both the ACC and CACC models are based on the collision-free IDM and were implemented using VISSIM’s driver behavior API. Figure 5.12 shows the driver behavior of vehicles that are already using CACC vehicle- following controls and how their lateral and longitudinal control is governed. It also shows the behavior of vehicles when they are joining a CACC string. These customized driver behav- ior algorithms are modeled as EDMs, and the switching between models is done based on contextual events through VISSIM’s COM capability. Although Lee’s model enabled dynamic switching between driver behavior models that replicated the behavior of lead and follower vehicles in a CACC string, additional enhancements were made to define string formation and dispersion mechanisms. These enhancements were implemented

Analysis and Evaluation Approach 81 based on recommendations from the California Partners for Advanced Transportation Technol- ogy, which is leading the CACC research on behalf of U.S.DOT’s Exploratory Advanced Research Program. The modified CACC algorithm included the following driver behavior models: • Preferential Lane Logic. When CAVs enter the system, their lane preference will be dynami- cally set to the left-most DL unless the static routing defines it to take the upcoming off-ramp. This logic is performed via VISSIM’s COM interface using the “Desired Lane” method. • String Formation Logic. This logic enables CAVs to form platoons based on their proxim- ity to each other. When strings are formed, the lead CAV (Leader) and the follower CAVs (Followers) get their respective EDMs assigned. For the Leader, this EDM enables it to act as an ACC vehicle. For the Followers, their respective EDMs enable them to act as CACC-equipped vehicles. EDM assignments are performed through VISSIM’s COM interface, whereas the actual vehicle behavior is controlled through EDM DLLs. • String Size Restriction Logic. When vehicles join existing CACC strings, they also can use the EDMs to check if the string has reached the maximum string size. This is achieved by intro- ducing a new vehicle type (in addition to Leader and Follower), called the String Terminator. Source: Lee et al. (2014). Start In Lane Change? In Platoon? Is Cut-In Join Allowed? Send Cut-In Join Request Signal to the Following Vehicle in the Target Lane Search Adjacent CACC Vehicles Identify the nearest CACC Vehicle (V) Is V in Same Lane? Is V in Downstream? Is V in Platoon? Is V in the tail of platoon? Is V in the middle of platoon? Set v=Leading Vehicle Set v=Target Vehicle for Lane Change Change Lane Set to “Lane Change Mode” Receive Cut-in Join Request? Ready to Join a Platoon Lane Changing to Join a Platoon C/ AV s a lre ad y in a p la to on Lo ng itu di na l C on tr ol in a P la to on Estimate Headway(h) h vs. Min Headway(H) h<H h>H h=H Collect Leading Vehicle Data (e.g., Speed, Acceleration, Distance) No No No Yes No No No No Yes Yes Yes Yes Yes Yes Record Simulation Results Decelerate (a) Accelerate (a)Maintain Speed Figure 5.12. CACC driver behavior logic.

82 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles The String Terminator behaves as a CACC-equipped vehicle but does not let another CAV join the platoon. This logic also is performed through VISSIM’s COM interface. • String Dispersion Logic. This logic actively checks for static routes of each vehicle in the CACC string. If a vehicle needs to exit through an upcoming ramp, it will split from the string. Leader-Follower-Terminator assignments are then reconfigured based on the new order of vehicles. This logic also is performed through VISSIM’s COM interface. 5.2.2 Dynamic Speed Harmonization The team used a simplified speed-based algorithm, developed by Ma et al. in 2016, which uses a simplified space-time relationship to approximate the typical complex models used by previous approaches. The vehicles upstream of congestion are provided a speed recommenda- tion that is a linear function of space (x) and temporal (t) speed measurements at appropriate intervals. The speed recommendation for a vehicle in space (x) and time (t) is given by: ( ) ( ) ( ) ( )= − ∆     +, ,s x t s t s t x x s t n m nm n where sn (t) represents the speed measurement at a point (n) at a time (t), and Dxnm is the distance between points n and m. After preliminary tests, this algorithm was modified to include a system that propagates the speed recommendations upstream to further reduce shockwaves and improve traffic flow. The team used speed measurement stations spaced at 0.1-mile increments and mapped vehicle locations to 0.1-mile resolution. The speed recommendations were updated every 15 seconds and provided to the equipped vehicles as their desired speeds. A minimum speed recommen- dation was kept at 25 mph, and the application was initialized only if a congested condition was detected on the DL. With these modifications, the final recommended speed for each vehicle space (x) and time (t) was: ( ) ( ) ( ) ( ) ( ) = + − ∆     +               , 25, 5, .s x t min s t s t s t x x s t mphn n m nm n A COM-based application was used to implement this DSH algorithm. The application watched for inputs from freeway sensors (data collection devices) for speeds every 15 seconds. When congestion was detected, the application calculated speed recommendations for every 0.1-mile sub-link. For each sub-link, desired speed points (similar to speed limit signs), were integrated into the network, which was updated based on the COM-based DSH application. Space-resolution of 0.1-mile was chosen based on the studies conducted by U.S.DOT under the DMA Impact Assessment Program to provide granularity. The update frequency of 15 seconds was chosen as a trade-off between the computation intensity of the algorithm and the travel time of vehicles on each sub-link. The overall implementation logic is shown in Figure 5.13. The DSH algorithm modeled in this project was implemented as a soft-control of the vehicle speeds in the sense that the vehicle controls assume the harmonized speeds as the new “desired speed.” As vehicles receive the harmonized speeds as desired speeds, each vehicle’s dynamics model tries to maintain a speed close to the received speed. In reality, vehicles using the DSH application might be paired with vehicles whose controls assume the new speed as a strict speed

Analysis and Evaluation Approach 83 control or with vehicles whose human drivers perceive the new speed as informational only and may choose not to follow the harmonized speed recommendations. 5.2.3 Combined Modeling of Applications The project team also modeled CACC and DSH applications in combination. Whereas the CACC application tends to reduce headways locally and increase lane capacity, the DSH applica- tion aims to increase lane throughput by providing early response to downstream congestion. Together, these applications could have significant mobility improvements on DLs. To test this hypothesis, evaluations were conducted by which CAVs could form close platoons when in the DLs. Additionally, the lead vehicle of each platoon was modeled as a DSH-equipped vehicle that receives speed recommendations based on any congestion detected downstream. Figure 5.14 demonstrates the two applications working in parallel on CAV DLs. Both the CACC and DSH applications were modeled in VISSIM using a software-in-the- loop system, as shown in Figure 5.15. The combined model included multiple internal and external driver behavior models and COM-based applications. The core of the system was an external simulation manager that controlled the VISSIM object at every time step. The simula- tion manager ran the simulations according to the scenario definitions in a scenario table and Figure 5.13. DSH algorithm implementation.

84 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles Figure 5.14. CACC and DSH working together on a platoon of CAVs. Scenario Table Performance Measures CACC Module DSH Module Simulation Manager MOE Logs Scenario Definitions Vissim Objects Reassign Driver-Behavior Models Extract Vehicle Data Extract Detector Data Override Desired Speed Figure 5.15. Overall evaluation framework.

Analysis and Evaluation Approach 85 exported the measures of effectiveness to a performance measures table. (The scenario table and performance measures table are discussed further in later sections of this chapter.). The CACC and DSH modules were used to manage CAV behavior and require constant interaction with the VISSIM platform. All of the interactions between the simulation manager and the VISSIM objects were handled via COM interface, but the CAV and non-CAV behavior was assigned as external and internal driver behavior modules. 5.3 Develop Scenarios for Assessment Once the overall evaluation framework had been developed, the next step was to define the specific scenarios to be tested. These scenarios were defined based on the research questions set forth in the analysis plan. This section discusses the various research questions that were addressed in the simulated scenarios. 5.3.1 Research Questions To support the development of guidance about conditions amenable to dedicating lanes to CAV users, the study team framed several research questions in the analysis plan. These questions were: 1. What will be the mobility, safety, and environmental benefits to users of DLs under different market penetrations of CAV applications: a. When they have exclusive access to these lanes? b. When they share the lanes with HOV/HOT vehicles? 2. What are the economic benefits to CAV DL users when compared to GPL users? 3. What are the incremental benefits of DSH application, when implemented with CACC application? 4. What are the impacts of the following factors on CAV DLs? a. Low versus high demand? b. Continuous versus restricted access to DLs? c. Temporary lane closures caused due to an incident? d. Slow-moving vehicle on a DL? 5.3.2 Simulated Scenarios To answer the research questions defined in the previous subsection, the project team devel- oped a list of simulated scenarios that could be used to provide performance measures. Each scenario was defined specifically to represent variations in the traffic mix, lane dedication, test- bed network, and other factors. The baseline simulation models were calibrated to the traffic mix identified in the correspond- ing case study site, and included HOVs and SOVs, with a percentage of heavy vehicles as per field conditions. For CAV test simulations, an overall percentage of vehicles was converted from normal vehicles to CAVs based on the assumed market penetration, with no increase in the actual demand. The MPR was used to represent the percentage of overall vehicles that are equipped with CAV applications such as CACC and DSH. Three lane dedication cases were considered in the simulations: • The base case allowed only HOVs on the DLs; • The priority lane case allowed HOVs and CAVs to share the DL; and • The exclusive lane case allowed only CAVs on the DLs. Table 5.2 depicts these three cases using red, yellow, and gray colors representing HOV, CAV, and GPL vehicles, respectively.

86 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles Table 5.3 shows a preliminary list of evaluation scenarios. In the table, each scenario is described in terms of the case study site, demand, market penetration, users allowed on the DLs, number of DLs, applications modeled, and any special cases. Scenarios with an N/A on the DL column represent locations with a higher MPR for which dedicating lanes would not be feasible. The baseline runs were calibrated to typical evening peak traffic conditions, and low and high demand conditions represented reductions and increases in the peak traffic by 20%. 5.4 Measure Performance of CAV Applications During the simulations, the project team logged a variety of performance measures to under- stand how dedicating lanes can impact both network performance and individual user needs in terms of efficiency. Four categories of performance measures were logged or computed from each simulation: mobility, safety, environmental, and societal equity. 5.4.1 Mobility Performance Measures The following mobility measures were logged from the simulation: • Average Vehicle Travel Time. VISSIM microsimulation can generate travel time performance measure without further modifications. A measurement of individual travel times for each vehicle across the simulation network can be refined into the individual vehicle classifications. The measurement of travel time per vehicle or average vehicle delay represent the vehicles’ “experiences.” By comparing the average travel time of vehicles between different simulations, the distortion due to variations in trip lengths can be normalized. • Average Travel Speed. Vehicle speeds can be quantified in several ways. For this project, the team used average network speed, which is quantified as the ratio of VMT and vehicle-hours traveled (VHT), to account for the variability in trip distances, travel segments, vehicle char- acteristics, and so forth. • Throughput. This performance measure was used to assess whether the DLs for CAVs increased the capacity of freeways by demonstrating whether more vehicles were able to pass through sections of the case study sites. In the scenarios modeled, the throughput was mea- sured between consecutive on-ramps and off-ramps using cordon lines. In the simulations for this report, the throughput was demonstrated by the percentage variation in cumulative number of vehicles passing through the freeway from the base case. The team also measured throughput per lane to distinguish the impact of CAVs on DLs and GPLs. T3 Scenario Description Depiction Base Case HOVs Only on the DL Priority Lane HOVs and CAVs Share the DL Exclusive Lane CAVs Only on the DL Table 5.2. Three cases of DLs.

Analysis and Evaluation Approach 87 ID Site Demand MPR (%) DL Users CAV Applications Special Cases 1 I-66 Typical p.m. Peak 0 HOV Base None 2 I-66 Typical p.m. Peak 0 HOV Base Lane Closure 3 I-66 Low Demand 0 HOV Base None 4 I-66 High Demand 0 HOV Base None 5 I-66 Typical p.m. Peak 10 CAV CACC None 6 I-66 Typical p.m. Peak 25 CAV CACC None 7 I-66 Typical p.m. Peak 35 CAV CACC None 8 I-66 Typical p.m. Peak 45 CAV CACC None 9 I-66 Typical p.m. Peak 100 N/A CACC None 10 I-66 Typical p.m. Peak 10 CAV+HOV CACC None 11 I-66 Low Demand 10 CAV+HOV CACC None 12 I-66 High Demand 10 CAV+HOV CACC None 13 I-66 Typical p.m. Peak 10 CAV CACC Lane Closure 14 I-66 Typical p.m. Peak 10 CAV+HOV CACC Lane Closure 15 I-66 Typical p.m. Peak 10 CAV DSH None 16 I-66 Typical p.m. Peak 25 CAV DSH None 17 I-66 Typical p.m. Peak 50 N/A DSH None 18 I-66 Typical p.m. Peak 100 N/A DSH None 19 I-66 Typical p.m. Peak 10 CAV + HOV DSH None 20 I-66 Low Demand 10 CAV + HOV DSH None 21 I-66 High Demand 10 CAV + HOV DSH None 22 I-66 Typical p.m. Peak 10 CAV DSH Lane Closure 23 I-66 Typical p.m. Peak 10 CAV + HOV DSH Lane Closure 24 I-66 Typical p.m. Peak 10 CAV DSH + CACC None 25 I-66 Typical p.m. Peak 25 CAV DSH + CACC None 26 I-66 Typical p.m. Peak 50 N/A DSH + CACC None 27 I-66 Typical p.m. Peak 25 CAV CACC Moving Bottleneck 28 I-66 Typical p.m. Peak 0 HOV Base Moving Bottleneck 29 I-66 Typical p.m. Peak 10 CAV CACC Restricted Lane Access 30 I-66 Typical p.m. Peak 25 CAV CACC Restricted Lane Access 31 I-66 Typical p.m. Peak 25 CAV DSH + CACC Restricted Lane Access 32 US-101 High Demand 0 HOV Base None 33 US-101 High Demand 0 HOV Base Lane Closure 34 US-101 Typical p.m. Peak 0 HOV Base None 35 US-101 High Demand 10 CAV + HOV CACC None 36 US-101 High Demand 25 CAV CACC None 37 US-101 High Demand 50 N/A CACC None 38 US-101 Typical p.m. Peak 10 CAV + HOV CACC None 39 US-101 Typical p.m. Peak 25 CAV CACC None 40 US-101 Typical p.m. Peak 50 N/A CACC None 41 US-101 Typical p.m. Peak 100 N/A CACC None 42 US-101 High Demand 10 CAV + HOV CACC Lane Closure Table 5.3. Preliminary list of evaluation scenarios.

88 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles 5.4.2 Safety Performance Measures Safety-based performance measures can contribute significantly to the development of guidance for dedicating lanes to CAVs. Specifically, they can help assess safety features that are required in the infrastructure when lanes are dedicated to CAVs under different conditions. For this study, two types of safety performance measures were captured in the simulation: • Lane Friction. This measure is defined as the difference in travel speeds of vehicles on the DLs and GPLs. Larger lane friction (a higher speed differential between these lanes) can render lane changes into and out of the DLs unsafe and hence warrants restricted lane access or physi- cal barriers/lateral spacing between DLs and GPLs. • Shockwaves and Speed Differential. Shockwaves and speed differentials occur due to changes in capacity, and they result in either static or moving bottlenecks along a facility. The properties of the shockwaves and speed differentials experienced by individual vehicles can be analyzed by studying the speed profiles of vehicles on a spatio-temporal scale (Dowling et al. 2015). The project team captured two types of speed differentials from the simulation: (a) spatial speed differential and (b) temporal speed differential. Spatial speed differential represents shockwaves. The speed differential is quantified by calculating the speed difference between adjacent segments for each time period and taking the 95th percentile throughout the simulation. The temporal speed differential is measured by calculating the speed difference between consecutive time periods for each segment and taking the 95th percentile throughout the simulation. A segment length of 0.1 mile and a time period of 15 seconds was utilized for these measures. The average of spot speeds within a segment was used to compute the speed differential, and the speed differential was compared between samples of simulation runs. 5.4.3 Environmental Performance Measures The research hypothesis was that both CAVs and non-CAVs would benefit from a more stable and laminar traffic flow in which the reduction in the number of vehicles accelerating and decel- erating across the network directly correlates with a reduction in fuel consumption and vehicle emissions. To test this hypothesis, the team coded environmental performance measures in the VISSIM microsimulation. Using data collected at defined nodes in the network, the model generated specific energy and emissions indicators based on individual vehicle trajectories con- sisting of speeds and accelerations. The four performance measures generated were carbon mon- oxide (CO), nitrogen oxides (NOx), volatile organic compounds (VOC), and fuel consumption. These measures were generated using the VERSIT+ model, which uses statistical models for detailed vehicle categories that are calibrated based on drive cycles (Smit et al. 2007). By utilizing instantaneous vehicle trajectories, the model also represents variations in fuel consumption as a function of speed and acceleration. 5.4.4 Societal Equity Performance Measures To distinguish the economic benefits to CAV DL users, the project team used value of travel time savings (VTTS) as the performance measure (U.S.DOT 2014). VTTS is a complex concept that expresses three principles: • Time saved from travel could be dedicated to production, yielding a monetary benefit to either travelers or their employers; • Time saved could be spent in recreation or other enjoyable or necessary leisure activities for which individuals are willing to pay; and

Analysis and Evaluation Approach 89 • The conditions of travel during part or all of a trip may be unpleasant for the travelers and involve tension, fatigue, or discomfort. For this study, the project team utilized average value of VTTS based on personal and business trips as per 2014 guidance. The VTTS was compared for DL CAV users and GPL users to distinguish between the economic impacts on both categories of users. 5.5 Analysis Assumptions Effectively modeling CAVs to assess their impacts on the transportation network and study the sensitivity of parameters is a complex task given the suitability of today’s microsimulation and other analytical tools. Hence, the analysis described in this report and project makes several assumptions to effectively model the behavior and performance of CAVs. This section describes some of the key assumptions used in this analysis. 5.5.1 CAV Application Modeling Assumptions Two CAV applications were modeled in this project—the CACC application and the DSH application. It should be noted that these applications are not standardized in any way, and other algorithms are available to model these applications and the variations within them. In addi- tion, these applications are still in the pilot deployment and testing phase, and they are not yet readily implemented in the market. Therefore, for purposes of this project, the research team employed certain assumptions regarding these applications. For example, the team assumed a maximum of five vehicles within a CACC string. This number was chosen based on some initial parametric testing aimed at considering the optimum number of vehicles that could platoon without significant string stability issues. Based on the performance measure used, other studies have shown varying values for this optimum string size. Similarly, the DSH was modeled to have a speed resolution of 5 mph. This and other parametric assumptions may have significant impact on the findings. The project team also assumed a simplified human-to-machine control transfer, because a full-fledged human factors study was beyond the scope of this project. 5.5.2 Demand and Capacity Assumptions Previous studies have indicated that CACC application is likely to increase the capacity of lanes and therefore may be able to carry a higher demand through the network. Consequently, the research team elected not to put any constraints in lane capacity. Consequently, the team was able to achieve higher lane capacities when CACC was used exclusively on DLs. This higher capacity may indirectly increase the demand on the network, however, because vehicles from other parallel corridors might switch routes to maintain equilibrium. In addition, the additional lane capacity might also increase the overall network VMT, latent demand, and other indirect factors. Given that the models used in this network were microscopic, the research team did not increase the demand on the network as an indirect consequence. To assess this impact fully, the model would need to include a software-in-the-loop system with an activity-based model for the entire region, which was beyond the scope of this project. 5.5.3 Performance Measurement Four types of performance measures were considered in this project: mobility, safety, envi- ronmental, and societal equity. The performance measures considered in each category reflect

90 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles only the direct performance of the network. For example, the team did not consider impacts due to increased VMT that might be caused by the increase in mobility performance of the net- work. The team also used simplified performance measures to reflect safety, environmental, and societal equity. Due to software restrictions, the analysis did not consider safety impacts such as changes in surrogate safety measures or direct crash indicators. For environmental performance, the research team did not calibrate the environmental performance models to the vehicle line-up that is reflective of the two regions modeled. Rather the team used the default model parameters. Societal equity was measured using VTTS. These calculation assumptions are discussed in more detail in Chapter 4. Long-term impacts on land use, economic measures, tolling, and other con- siderations were not part of this study. 5.5.4 Dedicated Lane/Geometric Assumptions Regarding the DL configuration and geometry the study team made several assumptions to distinguish the modeling scenarios clearly. For example, the geometry of DLs was assumed to be the same for HOV and human-driven lanes in terms of lane width, curvature, access and egress distances, and other details. In reality, some of these factors could be changed or reduced based on the robustness of applications and types of vehicles allowed. Similarly, the study team used strict modeling assumptions on vehicle types and access rules for the DLs, except in certain hypothetical scenarios. For example, when CAVs had exclusive access to DLs, it was assumed that no other vehicles would enter the DLs. The models also reflected tactical driving rules while accessing DLs. For example, CAVs that had shorter travel distances on highways did not access DLs for such short distances, as would happen in real life. As much as possible, the project team made realistic assumptions in modeling and developing scenarios to perform the evaluation within resources while answering several research questions that could inform the final guidance.