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.

20 This chapter summarizes the process used to select the CAV applications and evaluate the available modeling frameworks for use in this project. 3.1 Selection of CAV Applications Given that the objective of this research was to evaluate the conditions amenable for dedicat- ing lanes to CAVs, the team evaluated the various CV applications that have been envisioned by the U.S.DOT. Table 3.1 lists the CV applications and their suitability in being modeled as a CAV applica- tion in a dedicated freeway lane facility. The three criteria utilized in this selection process were: • Suitability to DLs, in the sense that the applications should support CAVs throughout their travel on a freeway corridor; • Suitability to the CAV environment (to eliminate CV applications that rely on human driving behavior); and • Adaptability to simulation models, so that their sensitivity and impacts could be assessed through modeling and simulation. The team evaluated 17 freeway applications. For informational purposes, Table 3.1 also includes the non-freeway applications on the U.S.DOT’s list of applications. Based on Table 3.1, the following nine applications are suitable for modeling as a CAV appli- cation within a freeway DL environment. • Application 1: Reduced Speed/Work Zone Warning. This application utilizes I2V communi- cation to broadcast alerts to drivers or vehicle control systems, warning them to reduce speed, change lanes, or come to a stop within work zones. In a CAV environment, this application uti- lizes the connectivity part to receive localized speed restrictions or closure information to adapt the desired speed and lane choice. Owing to its limited applicability (based on vicinity of work- zones), however, the team did not recommend modeling this application in a DL environment. • Application 2: Connected Eco Driving. This application uses V2V/V2I data to provide cus- tomized real-time driving advice to drivers as well as vehicle control systems, including recom- mended driving speeds and optimal acceleration/deceleration profiles, so that they can adjust their driving behavior to save fuel and reduce emissions. In a CAV environment, this application would be similar to platooning vehicles using an energy/emissions-optimized CACC system. • Application 3: Eco-Lanes Management. This application establishes parameters and defines the operations of eco lanes. Eco lanes are managed lanes that use traffic management strategies or enforce in-vehicle applications that aim at reducing energy consumption or emissions by the vehicles. Eco lanes require other applications, such as connected eco driving or eco speed C H A P T E R 3 Connected and Automated Vehicle Applications

Connected and Automated Vehicle Applications 21 harmonization, to be enforced to gain environmental benefits. In a CAV environment, an eco-lane would be a DL that provides exclusive access to vehicles controlled using an energy/ emissions-optimized CACC system or a speed harmonization system. • Application 4: Eco-Speed Harmonization (ESH). This application determines speed limits optimized for the environment based on traffic conditions, weather information, and green- house gas and criteria pollutant information, allowing for speed harmonization in appropriate areas. This application is similar to variable speed limit applications that are optimized for energy and emissions. In a CAV environment, the application would use V2I information to obtain dynamic speed limits and automated longitudinal control to govern the speed. • Application 5: Eco-CACC. This V2V application uses CV technologies to collect speed, accel- eration, and location information from other vehicles and integrate these data into a vehicle’s ACC system, thus allowing for automated longitudinal control capabilities and vehicle pla- tooning to reduce fuel consumption and emissions. This application is similar to the CACC application, but the speed-selection function utilizes energy/emissions modeling to minimize fuel use and emissions. • Application 6: DSH. This application aims to recommend target speeds to equipped vehicles in response to congestion, incidents, and road conditions to maximize throughput and reduce crashes. This application is similar to dynamic speed limits, where speed limits are governed to reduce sudden decelerations or shockwaves across the network. In a CAV environment, the application would use V2I information to obtain dynamic speed limits and automated longitudinal control to govern the speed. • Application 7: Queue Warning. This application aims to provide drivers with timely warn- ings of existing and impending queues. For purposes of CAV DL modeling, this application needs to be coupled with speed harmonization or route guidance applications. No. Application Title Suitability to Dedicating Lanes Suitability to CAV Environment Adaptability to Simulation Models 1 Curve Speed Warning N/A ü ü 2 Spot Weather Impact Warning N/A ü N/A 3 Reduced Speed/Work Zone Warning ü ü ü 4 Emergency Electronic Brake Lights ü ü N/A 5 Forward Collision Warning ü ü N/A 6 Blind Spot/Lane Change Warning N/A ü N/A 7 Connected Eco-Driving ü ü ü 8 Eco-Lanes Management ü ü ü 9 Eco-Speed Harmonization (ESH) ü ü ü 10 Eco-Cooperative Adaptive Cruise Control (Eco-CACC) ü ü ü 11 Eco-Ramp Metering N/A ü N/A 12 Motorist Advisories and Warnings N/A ü N/A 13 Dynamic Speed Harmonization (DSH) ü ü ü 14 Queue Warning ü ü ü 15 Cooperative Adaptive Cruise Control (CACC) ü ü ü 16 Incident Scene Pre-Arrival Staging Guidance for Emergency Vehicles N/A ü N/A 17 Incident Scene Work Zone Alerts for Drivers and Workers ü ü ü Table 3.1. List of CV applications and their applicability in a CAV environment.

22 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles • Application 8: CACC. This application aims to dynamically adjust and coordinate cruise con- trol speeds among platooning vehicles to improve traffic flow stability and increase through- put. CACC uses connectivity to link the speed-selection functions of vehicles in a platoon and automate control for precisely following a given speed profile. This application was already selected for modeling under this project. • Application 9: Incident Scene Work Zone Alerts for Drivers and Workers. This application warns on-scene workers of vehicles with trajectories or speeds that pose a high risk to their safety. It also warns drivers passing an incident zone if they need to slow down, stop, or change lanes. The reduced speed/work zone warning application is a subset of this application. In a CAV environment, V2I information would be utilized for vehicles to receive advisory speed and lane closure information. Automated control would adapt to this precise speed change and engage automated lane-change for this application. Owing to its limited application (based on vicinity of work-zones/incident zones), however, the team did not recommend modeling it in a DL environment for this project. 3.1.1 Recommendations After careful review of these applications, the following applications were selected for imple- mentation in NCHRP Project 20-102(08). Each of these applications is representative of mul- tiple applications reviewed in the previous section. • Recommended Application 1: CACC. This application aims at platooning equipped vehicles on a lane by adjusting and coordinating vehicle cruising speeds and headways. In addition, the vehicles use connected and automated control to form CACC platoons via rear, front, and cut-in joining as more equipped vehicles are found within the vicinity of each other. This application was reviewed in detail in Shladover et al. (2015). For purposes of this study, Eco- CACC and Connected Eco-Driving are similar applications, except that they aim at reducing energy/emissions as opposed to improving throughput and safety. • Recommended Application 2: DSH. This application aims to harmonize vehicle speed on the freeway to minimize shockwaves and potentially improve system mobility by detecting congestion or queues downstream. In a connected environment, vehicles use V2I communication to trans- mit information about their traffic states to a central command, which finds the optimum speeds for the vehicles to travel in the upstream sections and uses V2I communication to provide this information back to the vehicles. Similarly, ESH aims to optimize vehicle speeds to reduce energy/ emissions of the vehicles in the network. DSH and ESH usually are coupled with Queue Warn- ing, which is used to detect downstream traffic states that are congested or queued. Depending on downstream traffic states, the Queue Warning application will engage a speed reduction strategy to prevent sudden deterioration of traffic states over the freeway networks, thereby reducing chances of shockwaves. For this project, the research team proposed using a version of the DSH application that aims to implement an advanced speed control strategy over the DLs. The application seeks to adjust vehicle speeds to maximize throughput through a bottleneck or gradually reduce the speed of traffic approaching a congestion queue to reduce secondary crashes at the end of the queue. 3.2 CACC Application The CACC is a CAV application that uses a combination of sensors and V2V communication to enable vehicles to adjust their speeds automatically in relation to the preceding vehicles in the same lane. CACC-equipped vehicles utilize connectivity and a range of sensors to increase situational awareness and engage automated methods of acceleration and deceleration, which are more accurate than human control. CACC-equipped vehicles dynamically and automati- cally coordinate cruise control speeds within groups of vehicles to increase traffic throughput

Connected and Automated Vehicle Applications 23 significantly (Figure 3.1). By tightly coordinating vehicle movements, vehicle headways can be significantly reduced, resulting in a higher vehicle density. The coordination also produces a smoothing of traffic flow, or an improvement in traffic flow stability. A CACC string is defined as a group of CACC-equipped vehicles that use connectivity and automated longitudinal control to act as a platoon of vehicles with short headways (Shladover et al. 2015). 3.2.1 Existing CACC Application Models Precisely modeling CAV applications such as CACC is vital in understanding the conditions ame- nable to dedicating lanes for CAVs using modeling and simulation. Additionally, the analysis of CAV technologies generally requires detailed, high-resolution data, and advanced functionalities to prop- erly deal with the uniqueness of CAVs. This analysis differs from traditional traffic analysis studies such as signal optimization traffic impact analysis. Specifically, connected and automated tech- nologies rely on sensor-based and CV data and need to be modeled into any assessment framework. This section reviews the state-of-the-art CAV modeling tools used to conduct the analysis of CAV DL strategies. Six models are included for in-depth review and represent the most advanced DL applications for CAV. They are: • A macroscopic model by Nikolos et al. (2015), referred to as Nikolos’ Model; • A hybrid CAV analysis framework proposed by the Volpe National Lab (Smith et al., 2016), referred to as the Volpe model; • An Aimsun-based CACC evaluation model by Shladover et al. (2012), referred to as the CACC-Aimsun model; • The MICroscopic model for Simulation of Intelligent Cruise control (MIXIC) by Van Arem et al. (2006), referred to as the MIXIC model; • The Flexible Agent-Based Simulator of Traffic (FAST) by Arnaout and Bowling (2011), referred to as the FAST model; and • A Vissim-based CACC DL analysis model by Lee et al. (2016b), referred to as the CACC-Vissim Model. With some modifications to the tool, each of the selected modeling tools could be implemented to conduct the analysis of CAV DL strategies. In this report, additional discussion in Chapter 4 presents additional discussion combining the results of the research team’s review with an evalu- ation of each modeling tool in relation to its applicability for analyzing CAV DL strategies. 3.2.1.1 Nikolos’ Model Nikolos et al. (2015) proposed a macroscopic approach to examine the impact of ACC and CACC by incorporating traffic dynamics into a gas-kinetic traffic (GKT) flow model initially developed by Ngoduy (2013). Such a macroscopic approach can only represent the density difference between manual driving, ACC, and CACC; it cannot represent the vehicle-following dynamics differences among these different modes of operation. The GKT model handles the behavior of a group of vehicles, specified by their location (X), speed (V), and desired speed (V0) at any instant (t) and phase space density (p). Put another way, the Figure 3.1. Illustration of CACC application for CAVs (Semsar-Kazerooni et al. 2016).

24 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles phase space density (p) represents the expected number of vehicles driving with speed (V) while having desired speed (V0) for a certain roadway segment at a specific location (X) and time (t). The GKT model describes the changes in p caused by both the inflow and outflow in the phase space and the interactions (such as gaps between consecutive vehicles) between the vehicle groups. Thus, the p can be changed depending on the significance of the interaction. Given the CACC environment that enables the exchange of real-time driving information, the gap between vehicles in the CACC group will decrease, resulting in increased phase space density. Nikolos et al. (2015) applied the GKT model to evaluations of ACC and CACC under mixed- traffic conditions. Two simulation scenarios were tested: (1) a 6-mile-long basic freeway segment with homogenous traffic and (2) an 18-mile-long freeway segment with an on-ramp. Figure 3.2 and Figure 3.3 show the spatio-temporal changes of densities in manually driven vehicles and Source: Delis et al. (2015). Figure 3.2. Density change in homogenous traffic condition: ACC (left) and CACC (right). Source: Delis et al. (2015). Figure 3.3. Density changes with an on-ramp: ACC (left) and CACC (right).

Connected and Automated Vehicle Applications 25 CACC-equipped vehicles for both scenarios. No validation efforts were made, but this model produced reasonable results as demonstrated by the density changes in these figures. The results for CACC showed fewer instances of traffic shockwave—and almost no instances of aggressive propagation of traffic shockwave—due to vehicle acceleration and deceleration trajectory when compared to trajectory results for ACC. The results for ACC showed several instances of traffic turbulence caused by fluctuating density due to queuing propagation. Considering the length of the test segment, the model appeared suitable for evaluating the impact of CAV technologies in a large-scale area. However, because it assumes 100% MPR for either ACC or CACC, Nikolos’ Model was challenging to work with in relation to effectiveness under various MPR scenarios. To overcome this challenge, Delis et al. (2015) had extended the model by incorporating distinctive values for the relaxation time parameter of the phase space density function depending on the market penetration. Figure 3.4 displays the changes of densities with 10% and 50% of CACC MPRs for the homogeneous traffic conditions and Figure 3.5 displays the changes for the on-ramp case (Delis et al. 2015). Source: Delis et al. (2015). Figure 3.4. Density change in homogenous traffic condition: 10% CACC (left) and 50% CACC (right). Source: Delis et al. (2015). Figure 3.5. Density change with an on-ramp: 10% CACC (left) and 50% CACC (right).

26 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles 3.2.1.2 Volpe CAV Analysis Framework Smith et al. (2016) have proposed an analytical modeling framework for assessing the benefits of AV operations. The proposed Volpe CAV Analysis Framework is a comprehensive approach for the quantitative assessment of the wide-ranging impacts of various automation scenarios (or levels). Given that these scenarios serve as inputs to the framework, the outputs of the frame- work are intended to help inform policy decisions. The framework is designed to facilitate the comparison of multiple scenarios—the degree of V2I and V2V connectivity and the different level of automation. At the time of the research for this study, the Volpe model was a work in progress and therefore was not available for modeling as part of this project; nonetheless, the project team wished to include discussion about this proposed analysis framework owing to its overarching capabilities in modeling components of a transportation system. The Volpe CAV Analysis Framework also incorporates several interrelated sub-models to assess the impacts in terms of safety, mobility, energy/environment, transportation system utilization, accessibility, land use, and economic analysis, as shown in Figure 3.6. Figure 3.6 illustrates the data flow among the sub-models in the framework. The outputs from the safety, land use, accessibility, regional mobility, and energy/environmental sub-models feed into the economic analysis sub-model. The safety and vehicle mobility sub-models feed each other. The output of the vehicle mobility model also feeds into the energy/environmental and regional mobility sub-models. The regional mobility sub-model feeds to the sub-models for the land use, accessibility, and transportation system usage models. There are feedback loops from the land use and accessibility models to the transportation system usage model and from there to the vehicle mobility and safety models. With the sub-models, the framework is designed to evaluate the following AV applications: • Collision Avoidance, • Traffic Jam Assistance, Figure 3.6. Sub-model structure of the Volpe CAV Analysis Framework. Source: Smith et al. (2016). Vehicle Corridor Sp ati al R es ol uti on Temporal Resolution Region Nation Seconds Safety Vehicle Mobility Regional Mobility Transportation System Usage Accessibility Land Use Economic Analysis Energy/ Environmental Years

Connected and Automated Vehicle Applications 27 • CACC, • Automated Platooning, and • Full Automation in a Controlled Environment. As shown in Figure 3.7, the framework is triggered from the safety and mobility sub-models. Once an application is selected, the safety sub-model estimates the likelihood of crash occurrence by considering the behavior of the driver/vehicle. Inputs to the safety sub-model include the ini- tial safety environment (e.g., position, driving condition), the attributes of the unequipped and equipped vehicles, including the vehicle itself (e.g., light vehicle versus truck), vehicle control (e.g., the level of automation that exists), and the driver (e.g., reaction time distribution). The safety sub-model estimates safety performance measures such as crash probability and severity, which feed into the next sub-models in Figure 3.6. In parallel, the vehicle mobility sub- model handles longitudinal and lateral maneuvers of individual vehicles (e.g., car following, lane changing, braking). The same inputs as the safety sub-model are applied for the vehicle mobility sub-model, as shown in Figure 3.7. Given the inputs, the vehicle mobility sub-model calculates the driving performance of individual vehicles such as speed, travel time, speed dif- ference, headway, and acceleration/deceleration rate. These performance measures are provided to the next sub-models to estimate link-wise and region-wise performance for mobility, safety, and environmental assessment. Economic analysis combines all the outputs produced by each sub-model. Source: Smith et al. (2016). Figure 3.7. Overall data flow diagram in the Volpe framework.

28 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles At publication of this report, the sub-model for vehicle mobility remained in the development stage, utilizing a microscopic traffic simulation platform. Employing a step-wise approach, the entire framework is designed to evaluate selected AV applications in a simple single-lane condi- tion and expand to more complex road networks. The research team suggests that, once it is com- plete, the proposed Volpe framework will be extremely useful to conduct large-scale evaluations for CAV applications; however, it will not be available for use within this project’s timeframe. 3.2.1.3 CACC-Aimsun Model Shladover et al. (2012) developed a microscopic simulation model to evaluate the perfor- mance of ACC and CACC on highway capacity under various market penetration conditions. Utilizing Aimsun (a commercial-off-the-shelf microscopic traffic simulator) with a Software Development Kit (SDK), the authors constructed an evaluation platform to handle the various scenarios reflected by the MPRs of ACC and CACC. In addition to ACC and CACC, the authors introduced a vehicle group to represent vehicles equipped with a vehicle awareness device (VAD) capable of providing adjacent equipped vehicles with real-time Here-I-Am (HIA) information. This information includes real-time position, speed, and heading information. Although the HIA vehicles are unable to conduct automated operation, the information disseminated from them will be helpful for the operation of CACC under low MPRs that could represent an early stage of CAV deployment. Using 10% increments for each HIA, ACC, and CACC driver group, the researchers evaluated 192 simulation scenarios to assess the impacts of different combinations of HIA, ACC, and CACC on roadway capacity (Shladover et al. 2012). Starting from the baseline case (i.e., 0% MPRs for HIA, ACC, CACC), the impacts of the HIA versus CACC case and the ACC versus CACC case, producing 81 scenarios for each case, were examined using a 4-mile-long single-lane hypotheti- cal freeway segment modeled in Aimsun. Significantly, to achieve all the flow measurement in a steady condition, the test segment was modeled with no on- and off-ramps, resulting in no lane changes. The longitudinal control dynamic models for both ACC and CACC were obtained from empirical data. The ACC car-following rules are proprietary to Nissan, and the CACC longitu- dinal control was customized from Bu et al. (2010). Both algorithms were simplified to be mod- eled in Aimsun using its Micro-SDK. The car-following behaviors for both manually driven and HIA vehicles were controlled by the NGSIM oversaturated freeway flow model developed by Yeo (2008) and Yeo et al. (2008), which are based on Newell’s linear model. The desired target time gap settings of the ACC or CACC-equipped vehicles were selected based on field test observations (Shladover et al. 2009, Nowakowski et al. 2014, and Shladover et al. 2014). The distributions of target time gap for ACC and CACC were as follows: • ACC: 31.1% at 2.2 seconds, 18.5% at 1.6 seconds, and 50.4% at 1.1 seconds; and • CACC: 12% at 1.1 seconds, 7% at 0.9 seconds, 24% at 0.7 seconds, and 57% at 0.6 seconds. The base-case condition consisted of all manually driven vehicles and was calibrated to produce a capacity of 2,018 vphpl by considering the potential disturbances in vehicle motions and the diversity of driver gap selections. For the ACC and manually driven vehicles, the MPRs of ACC had a negligible influence on the achievable capacity, resulting in the narrow range from 2,030 to 2,100 vphpl. For the CACC and manually driven vehicles, the capacity increased from nearly 2,000 vphpl to nearly 4,000 vphpl, as shown in Figure 3.8. In this experiment, the CACC mode was available only when the CACC-equipped vehicle was behind another CACC-equipped vehicle or a vehicle equipped with a VAD, with the exception of the first vehicle of the CACC group, which used ACC mode to follow an unequipped lead vehicle. Thus, the increase of capacity grew qua- dratically as the MPR increased, resulting in 3,970 vphpl of capacity at the 100% CACC rate. It was also discovered that introducing VAD devices provides real-time vehicular information to the following CACC-equipped vehicle, which enables increased capacity. Figure 3.9 shows the capacity

Connected and Automated Vehicle Applications 29 increase with respect to the combination of CACC and HIA on the left and the combination of CACC and ACC on the right (Shladover et al. 2012). More recently, the CACC-Aimsun model was extended to include multi-lane freeway scenarios with entry and exit ramp traffic, allowing for assessment of more complex traffic scenarios under a current FHWA Exploratory Advanced Research Program project (Shladover et al. 2018). That model was calibrated for the baseline all-manual-driving scenario by use of detailed archived traffic data from the SR-99 freeway corridor in the Sacramento region. 3.2.1.4 MIXIC Model MIXIC is a non-commercial microscopic traffic simulator developed by the Netherlands Organization for Applied Scientific Research (also known as Toegepast Natuurwetenschappelijk Onderzoek, or TNO) for the assessment of the impacts of CAV applications (Van Arem et al. 2006). With a simulation resolution of 0.1 second, the MIXIC model estimates various perfor- mance measures covering mobility (e.g., travel time, delay), safety (e.g., time to collision), and Source: Shladover et al. (2012). Figure 3.8. Capacity increase by CACC MPRs. Source: Shladover et al. (2012). Figure 3.9. Capacity increase by the combination of CACC and HIA (left) and CACC and ACC (right).

30 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles environmental impacts (e.g., exhaust-gas emission, noise, fuel consumption). Based on a modular structure, MIXIC is flexible to customize vehicle models for handling longitudinal and lateral maneuvers. Making use of this flexibility, the authors conducted an assessment of Advanced Driver Assistance Systems (ADAS) such as ACC, automated platooning, special lane for intelligent vehicles, cooperative following and merging, V2V communications (also known as CarTalk), and CACC (Van Arem et al. 1997). Van Arem et al. (2007) also used MIXIC to examine the traffic flow impact of CACC on a 3-mile-long four-lane highway with a lane drop in the downstream causing a bottleneck (Fig- ure 3.10). The lane drop corresponded to a mandatory lane change in the MIXIC traffic simu- lation model. When a mandatory lane change is carried out, the drivers turn off their CACC systems. Under the normal conditions maintained by MIXIC, once the mandatory lane change has been carried out, the CACC systems are turned back on. A lane drop makes it possible to measure the maximum traffic volume at different MPRs of CACC when the traffic volume on the link before the lane drop nears congested state. In addition, a number of experiments were conducted with a special lane for CACC-equipped vehicles to study whether this would lead to additional traffic flow benefits (Van Arem et al. 2007). Using the hypothetical corridor shown in Figure 3.10, mobility performance was measured to examine queue length and speed around the bottleneck and throughputs under various CACC MPRs (i.e., from 0% to 100% in increments of 20%) for a single heavy-traffic-volume case (i.e., 7,600 vph). The simulation study demonstrated that CACC produced up to 80% reduction in queue length and up to 10% improvement in average speeds under 100% MPR, although no significant throughput improvements were observed (Van Arem et al. 2007). The MIXIC Model enabled development of a simulation test bed for CACC, but because the case study was conducted using only a single traffic-volume case, it did not explore the effectiveness of CACC under varying factors such as MPR, target headways, and so forth. Figure 3.11 shows the results. 3.2.1.5 FAST Traffic Simulation Model FAST is an agent-based microscopic traffic simulation model extended from the two-lane microscopic traffic simulation model originally developed by Treiber (2016). Figure 3.12 pro- vides a snapshot of the user interface. FAST uses a microscopic simulation approach to mimic Source: Van Arem et al. (2007). Left lane is CACC lane Non-CACC vehicles need to leave CACC-lane (pre-warning distance 1350 m) Steady State Lane drop of left lane (pre-warning distance 1350 m) Left lane dropped No steady stage No steady State Upstream Downstream Link 1 Link 1 Link 3 Link 4 Link 5 Link 6 Lane Drop 500 m 500 m1000 m Figure 3.10. Layout of simulation segment.

Connected and Automated Vehicle Applications 31 Source: Van Arem et al. (2007). Figure 3.11. Queue length and speed improvement by CACC. Source: Trieber (2016). Figure 3.12. Snapshot of FAST.

32 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles the behavior of individual driver’s longitudinal and lateral driving maneuvers and macroscopic approach to handle collective dynamics of traffic flow (like density, flow, shockwaves) to analyze the traffic performance. It is also flexible in allowing simple calibration of various parameters to conduct differing scenarios depending on the values assigned to the parameters. Employing the Intelligent Driver Model as a primary car-following model and MOBIL (Minimizing Overall Braking Induced by Lane Changes) as a lane-changing model, FAST is well suited for simulating complex traffic patterns developing over time. It is also flexible in allowing simple calibration of various parameters to construct differing scenarios depending on the values assigned to the parameters. Using an agent-based modeling structure, FAST handles thousands of agents to rep- resent individual vehicles at the micro level while collecting measures of effectiveness (MOEs) at the macro level. Pursuing open-source policy, FAST allows modelers to customize and share the source codes with other modelers. Arnaout and Bowling (2011) investigated the performance of CACC by customizing FAST. For a hypothetical 3-mile, 4-lane hypothetical freeway network with an on-ramp, the authors assessed the benefits of CACC under five traffic-volume scenarios (4k, 5k, 6k, 7k and 8k vph) and six discrete MPR levels (0%, 20%, 40%, 60%, 80% and 100%) using throughput and average speed measures. Assuming 0.5 seconds of CACC headway and 0.8 to 1.0 seconds of non-CACC headways, they showed that CACC dramatically improves both roadway capacity (Figure 3.13) and speed (Figure 3.14), resulting in up to 140% increase in throughput and up to 180% improvement in average speed at 100% MPR. Although they pursued realistic traffic conditions to generalize CACC performance, the authors’ experiments were somewhat limited, as only a single fixed on-ramp traffic volume was used, no CACC-equipped vehicles from the on-ramp were assumed, and only one CACC headway case was explored. 3.2.1.6 CACC-Vissim PTV Vissim is a commercial, off-the-shelf, multimodal microscopic traffic simulation software in which each entity (e.g., car, train, pedestrian) is simulated individually. Vissim’s ability to isolate the behavior of each entity is one of the most crucial elements for CACC con- cept implementation as well as evaluation. Lee et al. (2014, 2016a) used a customized version of Source: Arnaout and Bowling (2011). Figure 3.13. Capacity changes by CACC MPRs.

Connected and Automated Vehicle Applications 33 Vissim to conduct evaluations of the impact of CACC DL on traffic flow. In their 2014 study, Lee et al. focused on DL uses for CACC-equipped vehicles, with a set of lane-changing mod- els for the interactions between the CACC lane and adjacent GPL. The simulation experiment was conducted using Vissim and its Application Programming Interface (API) and Component Object Model (COM) modules. For their 2016 research, Lee et al. developed a CACC add-on package made up of three major modules—a Vissim Network module, a Simulation Manager module, and a customized API module, as shown in Figure 3.15. A 13-mile freeway segment of I-66 in Northern Virginia was selected as the simulation test-bed for both evaluations (see Figure 3.16). The selected site is a major corridor in Northern Virginia with severe recurrent congestion. The a.m. peak hour is eastbound into Washington, D.C., and the p.m. peak is westbound out of the District. During the peak hours in both directions, the left-most lane is designated as HOV-only and hard shoulder running. The primary goal of the simulation was to examine the system-wide impacts of early deployment CACC DL strategies on roadway performance based on a variety of external factors, as follows (Lee et al. 2014, 2016a): • Overall Demand: 100% (base condition) to approximately 120%, with 5% increments; • CACC MPR: 0% (base condition) to approximately 60%, with 5% increments; • DL Use Strategy: Base (current HOV lane), Mix-Managed (HOV + CACC); and CACC- Dedicated (CACC-only); • CACC Inter-Vehicle Target Headway: 0.6 second, 0.8 second, and 1.0 second; • Critical Lane-Changing Headway for the Leading Vehicle; • Critical Lane-Changing Headway for the Following Vehicle; and • Critical Lane-Changing Speed Difference for Both Leading and Following Vehicles. The mixed managed lane use strategy (shown in Figure 3.17) clusters the CACC-equipped vehicles in one lane to create locally high market penetration during the early, low market pen- etration phase. Dedicated CACC lanes can be adopted when the number of CACC-equipped vehicles is sufficient to fully utilize a lane. The research by Lee et al. (2016a) also examined lane- changing activities. Lane-changing activities for a new CACC-equipped vehicle to join a CACC platoon were considered in the simulation model by employing trigger conditions calculated by combining headways and speed differences for the target vehicle in the CACC lane. Specifi- cally, the CACC simulations showed that when the headway for the leading and the following Source: Arnaout and Bowling (2011). Figure 3.14. Speed improvement by CACC MPRs.

34 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles Source: Lee et al. (2016a). Figure 3.15. CACC simulation tool architecture. Source: Lee et al. (2016a). Figure 3.16. Simulation network.

Connected and Automated Vehicle Applications 35 vehicles was less than the critical safety threshold, and the speed difference between the leading vehicle and the following vehicles in the target lane was lower than the threshold, a lane change was activated. Notably, the CACC-equipped vehicles’ lateral behavior was not calibrated; this is because no knowledge yet exists of how the human drivers of non-CACC-equipped vehicles will behave when joining or leaving CACC platoons. The key modeling objective of the research by Lee et al. was to evaluate the mobility benefits of multiple CACC early deployment strategies under low to medium market penetrations. One important question with respect to CACC early deployment is whether a DL should be used. Sev- eral relevant prior studies had shown the effectiveness of CACC under low MPRs, but these prior studies did not consider the real-life difficulties of doubling a lane’s flow rate, such as lane changes under speed differentials and short gaps, how to dissipate vehicles from the lane, and so forth. Lee et al. (2016a) used three types of car-following behavior for their 2016 analysis: 1. Vissim’s default car-following model (i.e., psycho-physical car-following) for non-CACC drivers; 2. The Intelligent Driver Model (IDM) for adaptive cruise control (ACC) vehicles to represent the lead vehicle of a CACC platoon; and 3. A customized IDM to manage CACC longitudinal maneuvers. Both the ACC and CACC models were based on the collision-free IDM and were imple- mented using Vissim’s driver-behavior API. Compared to other models, Lee’s Vissim-CACC model can assess the potential system-wide benefits, such as total travel time, total through- put, total delay, and average speed, under a wide range of traffic scenarios. As an example, Figure 3.18 shows the system-wide performance measures for up to 30% CACC MPR. In addi- tion to the network-wide performance evaluation, it also can evaluate the localized impacts of CACC, such as mainline travel time and headway distribution of a certain location for safety analysis, as displayed in Figure 3.19. 3.2.2 Evaluation of Existing Analytical Models The previous section discussed six approaches to modeling DLs for CACC application that were considered by the project team for use in this project. One major shortcoming observed in many of these models was the lack of availability of the source-code or API for implementation in a state-of-the-art modeling software. Among the six models, Nikolos’ Model and the MIXIC Source: Lee et al. (2016a). CACC Vehicle in platoon CACC Vehicle not in platoon yetHuman-Driven Vehicle Platoon_Headway Head_Headway Figure 3.17. CACC platooning.

36 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles Source: Lee et al. (2016a). Figure 3.18. CACC simulation results: network-wide performance. Source: Lee et al. (2016a). Figure 3.19. CACC simulation results: mainline performance.

Connected and Automated Vehicle Applications 37 Model were unavailable for use by the project team. Because the Volpe framework was under development, it was dropped from the evaluation. Employing an open-source architecture, the FAST model was available for use and appeared to be flexible in relation to customizing for CAV simulations, as had been demonstrated by Arnout and Bowling (2011); however, the maturity of the FAST model appeared insufficient to fulfill certain crucial requirements for the evaluation of CAV DL strategies, such as modeling DLs, scripting capabilities, as well as flexibility to address certain performance measures. APIs to two commercial microsimulation models were available to the project team: CACC-Aimsun and CACC-Vissim, and the research team selected these modeling tools for further evaluation for use in this project. The next sec- tion describes the research team’s evaluation of these two models to select the best base model using a specific set of criteria. 3.2.2.1 Evaluation Framework Precisely examining the suitability of existing modeling tools for analyzing the impact of CAV DL strategies depends on proper design of the evaluation measures. This section presents the evaluation measures utilized to determine the best modeling tools for conducting CAV DL analysis. The research team categorized the evaluation measures by considering the benefit/ disbenefit analysis discussed in Chapter 2. The evaluation measures for the existing modeling tools were categorized as follows: 1. Ability to create a variety of vehicle classes with ability to specify class-specific parameters such as: a. Passenger car class: SOV, HOV; b. Transit vehicle class: bus, tram, train, taxi; c. Heavy vehicle class: truck, trailers; d. CAV class: CV, AV; e. Geometrics of the vehicles (e.g., dimension); and f. Driver behavior parameters (vehicle following parameters, vehicle lane-changing parameters, etc.). 2. Ability to model various facility types for lane restriction scenarios relative to vehicle category for separated and non-separated CAV-only lanes, including: a. CAV DLs; b. HOV/HOT lanes; c. On/off ramps; d. Auxiliary lanes for merging/diverging activities; e. Toll plazas/booths; and f. Lane drops/bottlenecks. 3. Ability to model CAV driving maneuvers such as: a. Longitudinal maneuvers for CAVs; b. Lane-changing (lateral) maneuvers for CAVs; c. Platoon manipulations (join, leave, create); d. CAV operational malfunctions; and e. Shorter time gap selections (fewer cut-ins, possible drag reductions). 4. Flexibility to customize modeling tools through API/scripting, such as: a. Parameters necessary to quantify the benefits and disbenefits of dedicating lanes to CAVs that are not native to the modeling framework in the “off-the-shelf ” version of the software; b. Dynamic re-routing of AVs and/or CVs to the most efficient route for an Origin-Destination pair (application to future study); c. Roadway capacity as a variable function of the CACC proportion in each iteration;

38 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles d. Modeling incidents/accidents; e. Modeling dynamic lane-drops/bottleneck conditions; and f. Modeling wireless communications characteristics such as latency and losses. 5. Ability to handle a variety of network sizes, including: a. Segments; b. Corridors; c. Regional areas; and d. States. 6. Ability to generate various MOEs, including: a. Mobility: LOS, throughput, travel time, average speed, delay, density, queue length, travel time reliability, volume-to-capacity ratio, total travel distance, number of stops; b. Environment: Gas emissions, fuel consumption; c. Safety: Surrogate measures (e.g., time to collision), crash rate; and d. Other: Ridership, benefit-cost analysis. 7. Usability of modeling tools, including: a. Graphic display of modeling operations (e.g., animations); b. User-friendly interface (e.g., Graphical User Interface) for operation; c. Graphical and user-friendly network builder; d. Input data requirements (e.g., roadway geometry, origin/destination tables, turning designations); e. Computational time; f. Calibration efforts; g. Default values for model parameters; and h. Ability to integrate with other relevant software (e.g., GIS tools, database software, the MOtor Vehicle Emission Simulator [MOVES]). Using the evaluation measures in these seven categories, the two models were ranked for their suitability to model CAV applications in DLs. The project team used qualitative analysis to determine the availability of each evaluation measure. The following indicators were applied for the evaluation: • Fully Available (). This indicator means the subject modeling tool is able to provide a func- tionality required to conduct and/or handle the corresponding evaluation measure within the current capability of the modeling tool. • Highly Available ( ). This indicator means the subject modeling tool is able to perform and handle the evaluation measure with minor additional customization efforts. Minor customiza- tion efforts could include model parameter value adjustment, simple scripting, network enhance- ment, and additional data provision, which can be achieved without source code modification. • Limited Availability (). This indicator means the subject modeling tool is able to perform and handle the evaluation measure with major additional customization efforts. Major cus- tomization efforts could include source code modification, adding new models (e.g., car- following/lane-changing), and complex scripting if applicable in the tool. • Unavailable (). This indicator means the subject modeling tool requires a model change to incorporate the feature (e.g., macroscopic to microscopic). 3.2.2.2 Evaluation Results, by Category The project team ranked the two tools by assigning qualitative indicators (ranging from unavailable through fully available) identifying the availability of each feature in the categorized evaluation framework. This section summarizes the evaluation results, by category. 1. Ability to Create a Variety of Vehicle Classes. Vehicle classes were divided into SOV, HOV, transit, heavy vehicles, and CAVs, which have diverse characteristics such as length, height, width, and driving performance (e.g., acceleration/deceleration). As summarized in Table 3.2,

Connected and Automated Vehicle Applications 39 the Aimsun- and Vissim-based CACC modeling tools developed by Shladover et al. (2012) and Lee et al. (2016a), respectively, were capable of handling the various vehicles classes without requiring additional customization. 2. Ability to Model Various Facility Types. The ability to model various facility types is one of the most crucial elements for precisely assessing the impact of CAV DL strategies. Highway facilities such as on- and off-ramps, auxiliary lanes, and toll plazas and booths also affect the opera- tion of CAV DLs. Furthermore, lane drops and/or bottlenecks caused by various reasons (e.g., work zone, incident/accident) also directly affect the performance of CAV DL operation. Thus, a modeling tool for CAV DL analysis must handle a variety of roadway facility types. As summarized in Table 3.3, the Vissim-based CACC modeling tool can handle such a need within the current modeling functionality. The Aimsun-based CACC model developed by Shladover et al. (2012) can deal with the modeling requirement with minor modifications to the current modeling tool. 3. Ability to Model CAV Driving Maneuvers. As reviewed in Chapter 2, CAV driving maneuvers for longitudinal and lateral movements differ from those of human-driven vehicles. As summarized in Table 3.4, both the CACC-Aimsun and CACC-Vissim modeling tools are equipped with a longitudinal maneuver model for CAV and appear capable of handling CAV driving maneuvers. Notice that the CACC-Aimsun model has only been applied to a limited number of studies involving lane changing. 4. Flexibility to Customize Modeling Tools Through API Scripting. Unlike traditional traffic analyses (e.g., traffic impact analysis), analyzing the impact of CAV DL strategies likely requires advanced functionalities that are not native to off-the-shelf modeling tools. To conduct seamless analysis, the modeling tools should be able to provide some extent of modeling flex- ibility. Table 3.5 shows the evaluation results regarding the flexibility of the selected modeling Evaluation Measure: Ability to Create a Variety of Vehicle Classes with Ability to Specify Class-Specific Parameters CACC (Aimsun) CACC (Vissim) SOV HOV Transit Vehicle Class (e.g., bus, tram, train, taxi) Heavy Vehicle Class (e.g., truck, trailers) CAV Class Geometrics of the Vehicles (e.g., dimension) Separated Driving Behavior Parameters for Each Class : Unavailable; : Limited Availability; : Highly Available; : Fully Available Table 3.2. Evaluation results for ability to create variety of vehicle classes. Evaluation Measure: Ability to Model Various Facility Types CACC (Aimsun) CACC (Vissim) CAV DLs HOV/HOT Lanes On/Off Ramps Auxiliary Lanes for Merging/Diverging Activities Toll Plazas/Booths Lane Drops/Bottlenecks : Unavailable; : Limited Availability; : Highly Available; : Fully Available Table 3.3. Evaluation results for ability to model various facility types.

40 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles tools. Utilizing commercial-off-the-shelf products, Aimsun- and Vissim-based CACC mod- eling tools are fully capable of customizing recorded MOEs. However, the Aimsun-based modeling tool needs to be modified to deal with incident/accident modeling and dynamic lane-drop/bottleneck modeling. 5. Ability to Handle the Variety of Network Sizes. CAV DLs will be deployed on a multi- lane freeway corridor with multiple interchanges producing the wide variety of lane-changing activities. To some extent, such a corridor would have existing dedicated/exclusive lane facilities, such as an HOV/HOT lane with a barrier and HOV/HOT-dedicated access ramp. It is necessary to examine the collective impacts of CAV DL strategies on adjacent corridors. Furthermore, measuring region- and state-wide effectiveness of CAV DL strategies is also crucial for planning purposes. As shown in Table 3.6, the Aimsun- and Vissim-based modeling tools are unable to conduct region- and state-wide analysis owing to their microscopic nature. The Vissim-based CACC modeling tool, however, has been applied for evaluating the impact of a CACC lane on a single corridor with multiple interchanges. 6. Ability to Generate Various MOEs. Depending on the benefit/disbenefit categories examined, evaluations of CAV DL strategies can incorporate a wide spectrum of performance measures. Thus, producing proper performance measures is the most critical requirement for the modeling tools for CAV DL analysis. Table 3.7 summarizes the capabilities of each modeling tool to generate performance measures critical for assessing the benefits and disbenefits of CAV DLs. The MOEs summarized in Table 3.7 are quantitatively measurable; qualitative performance : Unavailable; : Limited Availability; : Highly Available; : Fully Available Evaluation Measure: Ability to Model CAV Driving Maneuver CACC (Aimsun) CACC (Vissim) Longitudinal Maneuver for CAV Lane-Changing Maneuver for CAV Platoon Formation and Dissolution Minimum and Maximum Size of the Platoon Shorter Time Gap Selections (Fewer Cut-Ins, Possible Drag Reductions) CAV Operational Malfunction Table 3.4. Evaluation results for ability to model CAV driving maneuver. : Unavailable; : Limited Availability; : Highly Available; : Fully Available Evaluation Measure: Flexibility to Customize Modeling Tools Through API Scripting CACC (Aimsun) CACC (Vissim) Parameters to Estimate MOEs That Are Not in the “Off-the-Shelf” Version of the Software Ability to Define the Capacity as a Variable Function of the CACC Proportion in Each Iteration Dynamic Re-Routing for CAV Modeling Incident/Accident Modeling Dynamic Lane-Drops/Bottleneck Conditions Wireless Communications Impact on Modeling CAVs Table 3.5. Evaluation results for flexibility to customize modeling tools through API scripting.

Connected and Automated Vehicle Applications 41 measures for societal justice such as equity and perception of exclusivity were excluded from the table. As seen in Table 3.7, almost all mobility measures are available through the selected modeling tools. Combining the Aimsun or Vissim tools with external environmental models such as MOVES (U.S. EPA 2016) and the VT-Micro Model (Rakha et al. 2004) could potentially enable estima- tion of the environmental impacts of CAV DL strategies. Obtaining safety measures appeared challenging, however, as neither Aimsun nor Vissim could generate the necessary number of crashes and crash severity. Methods such as the Safety Impact Methodology (SIM), developed by Carter et al. (2009), also could be used to enhance these tools to account for safety. The SIM is a systematic approach for evaluating the safety impacts of a new vehicle system by incorpo- rating historical crash, driver performance, and system performance data to enable a rigorous comparison of baseline and treatment vehicle crash conflicts. SIM has been used to assess the safety impacts of V2V technologies (Harding et al., 2014). As Harding et al. had used SIM in : Unavailable; : Limited Availability; : Highly Available; : Fully Available Evaluation Measure: Ability to Handle the Variety of Network Sizes CACC (Aimsun) CACC (Vissim) Segment (Without Interchanges) A Single Corridor Including Multiple Interchanges Multiple Corridors Region State Table 3.6. Evaluation results for ability to handle the variety of network sizes. : Unavailable; : Limited Availability; : Highly Available; : Fully Available Evaluation Measure: Ability to Generate Various Measures of Effectiveness CACC (Aimsun) CACC (Vissim) Mobility Level of Service Throughput Total Travel Time (VHT) Total Travel Distance (VMT) Average Speed Delay Queue Length Travel Time Reliability Environment Gas Emission (Carbon Monoxide, Carbon Dioxide, Nitrogen Oxides, Hydrocarbons) Fuel Consumption Safety Surrogate Measure (e.g., Time to Collision) Speed Difference (Delta V) Number of Crashes Crash Severity Other Ridership Route Diversion Mode Diversion Table 3.7. Evaluation results for ability to generate various MOEs.

42 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles assessing safety impacts, the NCHRP project team could use the Surrogate Safety Assessment Model (FHWA 2008a) in this study by integrating it with the Aimsun or VISSIM modeling tools to enable estimates of surrogate measures such as time to collision, post encroachment time, and speed difference (delta V). Other measures, mainly required for planning purposes, were unavailable from the selected modeling tools. 7. Usability of Modeling Tool. From the perspective of an analyst, the usability of a model- ing tool is a critical element significantly affecting the efficiency of modeling activities. Table 3.7 summarizes the usability requirements commonly acceptable for the practice of transportation modeling. Based on commercial-off-the-shelf product platforms, the Aimsun- and Vissim-based modeling tools provide the highest usability, as shown in the table. In Table 3.8, the input data requirement, computational time, and calibration efforts are not ranked using an Unavailable () to Fully Available () scale, but rather using a Low, Moderate, and High scale. 3.2.2.3 Overall Evaluation Results Table 3.9 summarizes the overall evaluation results for the two modeling tools based on their usability scores to the evaluation measures established. Owing to high scores in this evaluation and availability as an open-source code in the U.S.DOT’s Open Source Application Develop- ment Portal, the project team suggested selecting CACC-Vissim as the selected modeling tool, with minor modifications. A modeling approach was proposed to use the CACC-Vissim model with enhancements related to DSH. Modifications required to execute the scope of this project are discussed in Chapter 5. 3.3 DSH Application DSH is another CAV application that is mature in terms of prototype development and field testing and can be implemented in a freeway DL environment. Several algorithms exist, but the general objective of the application is to harmonize the speeds of vehicles upstream on a freeway Evaluation Measure: Usability of Modeling Tool CACC (Aimsun) CACC (Vissim) Graphically Display Modeling Operations (e.g., Animation) User-Friendly Interface for Operation (e.g., GUI) Graphical and User-Friendly Network Builder Input Data Requirement High* High* Computational Time High* High* Calibration Efforts (Network and Demand) High* High* Default Value Availability for Model Parameters Integration with Other Relevant Software *Depending on network size; : Unavailable; : Limited Availability; : Highly Available; : Fully Available Table 3.8. Evaluation results for usability of modeling tool. Unavailable Limited Availability Highly Available Fully Available CACC (Aimsun) 7 0 19 26 CACC (Vissim) 7 0 6 39 Table 3.9. Overall evaluation results.

Connected and Automated Vehicle Applications 43 to minimize shockwaves and potentially improve system mobility by detecting congestion or queues downstream (Ma et al. 2016). Figure 3.20 demonstrates the process: once the traffic management center (TMC) identifies congestion in a freeway segment, the speed harmonization application will compute speed recommendations for upstream freeway segments to enhance the throughput and avoid sudden deceleration and braking (thereby reducing shockwaves and the probability of secondary collisions). In a connected environment, the vehicles use V2I communication to inform a TMC of their traffic states. When vehicles slow at a bottleneck, the TMC identifies their V2I communication as a signal of impending congestion. The TMC selects the optimum speeds for vehicles traveling upstream and uses I2V communication to relay this speed information to the upstream vehicles. The upstream CAVs receive and implement the recommended speeds, which has the effect of reducing congestion and attendant risks. 3.3.1 Existing DSH Application Models The general objective of DSH is to smooth traffic speeds on a freeway in both temporal and spatial dimensions. The application functions as a derivative of variable (or dynamic) speed limits. Several algorithms exist for dynamic speed limit strategies, and many of them have been modified to provide a more granular traffic-smoothing strategy for DSH implementation using CV technology. Table 3.10 provides a summary of the different speed harmonization applica- tions previously researched. 3.3.2 Modeling the DSH Application As a starting point in modeling the DSH application, the team used a simplified speed-based algorithm developed by Ma et al. in 2016. This algorithm uses a simplified space-time relation- ship to approximate the typical complex models used by previous approaches. The vehicles upstream of congestion were provided a speed recommendation that is a linear function of spatial (x) and temporal (t) speed measurements at appropriate intervals. The speed recom- mendation 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 in space (n) at a specific time (t), and Dxnm is the distance between point (n) and a second, downstream point (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. Figure 3.20. DSH process flow.

44 Dedicating Lanes for Priority or Exclusive Use by Connected and Automated Vehicles Study CAV/CAV V2I or V2V Control Algorithms Lu et al. (2005) CV V2I Application aimed at reducing the speed limits of freeway segments upstream of a bottleneck in proportion to the observed bottleneck speed if vehicle flow throughputs are above the bottleneck capacity. Talebpour et al. (2013) CV V2I A wavelet-transform based algorithm was used to detect formation of perturbations, and a cognitive risk-based microscopic simulation model was adopted to account for human behavior. A reactive speed limit was selected to implement speed harmonization. INFLO Project (Dowling et al. 2015) CV V2V, V2I The algorithm grouped freeway sub-links with similar recommended speeds to produce harmonized speeds, which was calculated as the speed of the slowest vehicle or within 5 mph interval of the downstream sub-link. Li et al. (2014) CAV V2V Used a CAV car-following rule that effectively suppressed the development of oscillations and consequently mitigated fuel consumption and emissions. Ma et al. (2016) CAV V2I Used a simplified space-time relationship to reduce traffic state oscillations to enhance traffic flow. Wang et al. (2015) CAV V2I Used aggregated traffic state information to detect the formation of congestion at a bottleneck, and each CAV processes the VSL signals from the central control unit individually. Yang and Jin (2014) CV V2V Advisory speed limit is calculated by each individual vehicle and then averaged among equipped vehicles. Ahn et al. (2013) CAV V2V Used a rolling horizon-based optimization approach to control vehicle speed within a preset speed window in a fuel-saving manner. Table 3.10. Review of previous speed harmonization algorithms (adapted from Ma et al. 2016). The team used speed measurement stations at 0.1-mile spacing and mapped vehicle locations to 0.1-mile resolution. The speed recommendations were updated every 15 seconds and pro- vided to the equipped vehicles as their desired speeds. A minimum speed recommendation 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

Connected and Automated Vehicle Applications 45 A COM-based application was used to implement this DSH algorithm. The application watched for inputs from freeway sensors (data collection devices) that recorded speeds every 15 seconds. When congestion was detected, the application calculated speed recommenda- tions for every sub-link of 0.1-mile length. For each sub-link, desired speed points (similar to speed limit signs) were integrated into the network, which was updated every 15 seconds via the COM-based DSH application. To provide granularity, space-resolution of 0.1-mile was chosen based on the studies conducted by the U.S.DOT under the Dynamic Mobility Applica- tions Program’s Impact Assessment (Dowling et al. 2015). 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 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 harmonized speeds as desired speeds, their vehicle dynamics model makes adjustments to maintain the vehicle speeds close to the received speed. In reality, the DSH application might be paired with a combination of vehicle controls such that some vehi- cles could assume the new speed as their strict speed control while, in other vehicles, human drivers might perceive them as informational only and may not follow the harmonized speed recommendations.