Framework for Assessing Potential Safety Impacts of Automated Driving Systems (2022)

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.

15   The framework will help state and local agencies follow a systematic methodology to assess the impact of specific automated driving system (ADS) applications on transportation safety. The steps and key elements of the framework are provided in Figure 6. As shown in Figure 6, the framework starts with identifying the ADS features that agencies would like to assess. The second step is to understand different aspects of the ADS feature such as functionality and expected market penetration rates. The developer describes the route and physical and environmental boundaries within which a particular function is designed to work. In other words, the developer specifies the operational design domain (ODD) of the feature. This ODD defines the deployment scenarios in the third step. Once the deployment scenarios are C H A P T E R 3 Overview of the Framework Elements Figure 6. Framework to assess safety impact of ADS (Source: Booz Allen).

16 Framework for Assessing Potential Safety Impacts of Automated Driving Systems defined, it is important to understand the technology and infrastructure dependencies. This helps to understand the risks and opportunities involved. With the scenarios identified, the fourth step is to define safety goals and the associated hypotheses. Step 5 involves identifying data sources and data metrics to help evaluate the stated hypotheses. The analysis method is also chosen during this step to derive insights from the data. It is important that the framework inherits feedback loops so that it can be refined as new information becomes available and technology advances. The framework should also follow an iterative approach as safety goals, questions, and hypotheses could change throughout the proj­ ect. To this end, the framework could go through iterations of refinement or feedback loops, as shown in Figure 6. The input for refining the framework depends on the output of Steps 5 and 6. These refinement iterations entail updating the framework at Steps 2, 3, 4, or 5—and consequent steps—depending on the type of new information. For instance, if new information became avail­ able regarding the expected market of an ADS feature, the refinement would include Step 2 and all consequent steps. However, if new advances in the ADS technology occurred and were anticipated to expand the feature’s ODD, the framework should be updated at Step 3 and all following steps. Finally, upon completion of Step 5, the results are communicated to internal and external stakeholders involved in the connected and automated vehicle ecosystem. It is important to share the safety impact of the ADS feature with all involved stakeholders and obtain feedback as much as possible to support decision­making. Step 1—Select ADS Feature to Assess Today, both technology companies and automotive manufacturers are investing heavily to develop automated features for vehicles. These technologies are expected to help transporta­ tion system users meet their needs and help agencies achieve their goals. The ADS performs the dynamic driving task (DDT) instead of a human and can reduce human error to improve safety. The driver can focus on other activities during commutes. These automated vehicle (AV) technologies are advancing every day and continue to transform the transportation landscape. A wide range of entities in both the public and private sectors are playing instrumental roles in advancing and implementing AV technologies, including transportation agencies, tech­ nology firms and service providers, major automakers, and research organizations. With AV technologies on the rise, state and local infrastructure owners and operators (IOOs) are seek­ ing new approaches to designing, constructing, operating, and maintaining highway infra­ structure. A variety of ADS features are out in the market to help meet user needs and to improve safety in the transportation network. IOOs may first identify and select the ADS feature of interest and understand its functionality and potential market penetration rates before going in depth to assess its safety impact. The framework provides the flexibility to assess one ADS feature at a time, or multiple ADS features together, to estimate the combined safety performance. To identify and select ADS features for safety analysis, IOOs can evaluate their local trans­ portation environment, including user needs and local trends, state and municipal policy and regulations, development and population densities, and local service providers. All of these factors will determine which ADS features enter the market. Stakeholder engagement throughout this step can be beneficial and may help to understand better public opinion, perception, and preference. ADS features in public transportation will likely disrupt urban markets while single­ passenger vehicle ADS features will likely be prevalent in commuting suburban communities. Practitioners should take care in evaluating relevant features to maximize both safety and mobility benefits of ADS.

Overview of the Framework Elements 17   Step 2—Understand the ADS Feature Description of ADS Feature The best sources for understanding the functionality of the ADS feature are the original equip­ ment manufacturer’s user manuals and online technology reviews. These sources not only pro­ vide a high­level description of the feature’s functions but can help provide context in terms of technology and infrastructure dependencies. Manuals will help IOOs understand the ODD and assist with defining deployment scenarios. By understanding the ADS feature performance, for example, by sensor suite, IOOs are able to derive performance capability of incoming ADS and validate infrastructure requirement gaps to accommodate minimal thresholds of sensors. For example, if a Level 2 vehicle is equipped only with cameras and radar, then the IOO in ques­ tion may want to understand the perception capability of the known equipment and identify whether both a human user and Level 2 vehicle can properly identify and react to roadway signs or markings. Expected Market Individual ADS features are unique, with disruption rates varying by location and transporta­ tion market segment. Understanding penetration trends in specific deployment contexts is helpful to estimate scale and timeframe of safety impacts in an area. Assessing how the technology will deploy over time is imperative for practitioners because impacts are often not linear with market penetration rate. ADS adoption will likely be uneven, both geographically and temporally, and individual jurisdictions may experience both extremes (i.e., highly automated vehicle fleets in one community and mostly traditional vehicles in another) in the adoption of ADS simultaneously. The uneven geographic deployments could be attributed to the geographical inconsistencies of many factors—at state and local levels—including customer acceptance, willingness to pay, policy and regulation, and willingness to share. Furthermore, this multispeed adoption scenario will affect interagency planning, particularly as funding at a state level is allocated to local priorities. Envisioned ADS Market, Fleet, and Vehicle Miles Traveled (VMT) The deployment future of ADS technology or automation level is speculative and typically could follow one of two principal scenarios, namely, low and high disruption scenarios. Accordingly, to account for more realistic scenario planning, this framework predicts two principal timelines— low­ and high­disruption scenarios—for the deployment of ADS­equipped vehicles. The key assumptions for these scenarios are outlined in Table 3 (McKinsey & Company, 2016). For the reasonable evaluation and selection of the scenarios, the penetration rates of ADS­ equipped vehicles were forecasted for a 10­year period. This range was based on the current Factors High Disruption Low Disruption Pace for addressing regulatory challenges Quick Slow Customer acceptance Enthusiastic Limited Willingness to pay Enthusiastic Limited Willingness to share trips Enthusiastic Limited ADS penetration More Less Table 3. Factors considered in the disruption scenarios.

18 Framework for Assessing Potential Safety Impacts of Automated Driving Systems understanding of the rapid advancement of technologies focused on Level 3 and 4 vehicle technologies that directly affect stakeholders. The predictions were extrapolated for both the high­ and low­disruption scenarios. Also, a saturation penetration rate of about 29% was assumed, based on surveys of customers’ willingness to pay discussed in Appendix A (Mosquet et al., 2015). This saturation value was used as a cap for expected market penetration of the pertinent features over the next 10 to 15 years. The proportion of shared highly automated vehicle (HAV) forecasts and the predictions from the McKinsey study (McKinsey & Company, 2016; Litman, 2019) were used to raise the predictions for the ADS­equipped vehicles at 2030. In this study, the price after launch was assumed to decrease by a compound annual rate of 4% for Level 3 and Level 4 features. Finally, recent reports and market forecasts specific to ADS features were used to scale the prediction values to be as consistent as possible with recent data. Table 4 outlines the envisioned ADS­equipped vehicles market and fleet shares and their proportion. More details about the statistics and numbers used to obtain these predictions can be found in Appendix A. Step 3—Define Deployment Scenarios Deployment scenarios ground the analysis in theory but incorporate realistic timelines that feature potential technical solutions situated within local context. These scenarios serve as the basis for identifying risks and opportunities for ADS technologies. The deployment of ADS technology may take many forms and occur on different timelines depending on a number of factors. This step helps to understand the potential technical solutions and deployment factors and how they relate to user needs and agency goals and processes. Specifically, Step 3 defines the penetration rate, ODD, and limitations of the ADS, which in turn help to identify the potential crashes impacted by ADS. For example, the ODD defines the geographic extent of crashes that could be impacted by the ADS (i.e., the ADS can impact only crashes on facilities over which it operates). The following subsections describe the use of penetration rates, ODD, and limitations of ADS in defining the deployment scenario. ADS Penetration Rates Having an estimate of the market penetration rates of ADS­equipped vehicles—and their equivalent proportion of the fleet—is an indispensable element of any planning effort for this technology. Benefits and disbenefits of ADS can vary significantly depending on the pro­ portion of ADS within the ODD. In other words, when both deployment rates of supporting infrastructure and penetration rates of ADS are high, the impacts may be more visible and quantifiable. Depending on the type of investment in the infrastructure, the envisioned benefits are expected to vary significantly based on the expected penetration rate of ADS technology. Year High Disruption Low Disruption Market Share Fleet Share VMT Share Market Share Fleet Share VMT Share 2020 1% <1% 1% 0% 0% 0% 2025 10% 7% 10% 2% 1% 1% 2030 30% 14% 22% 6% 3% 4% Table 4. ADS-equipped vehicles market penetration and fleet proportion predictions.

Overview of the Framework Elements 19   On the other hand, the future of the deployment rate of the ADS is speculative and could follow diverse scenarios since it hinges on several interrelated factors, such as reliability of tech­ nology, regulatory challenges, and consumer acceptance and willingness to pay. For instance, technology innovations in this area would continuously supersede first­generation technologies by more advanced systems. In such cases, the suitability of ODDs may change and enable more expansive domains (such as active operations in work zones or adverse weather), which may affect the ADS deployment rate. To account for such a speculative future, state and local agencies typically follow a scenario planning approach to define the set of possible futures and corresponding policy responses that support the community vision and goals. In this approach, planners typically envision multiple scenarios with different assumptions regarding factors such as reliability of technology, regulatory challenges, consumer acceptance and willingness to pay, and private ADS ownership, among others. As a result, this framework first analyzes a synthesis of published reports (both manufacturer­specific and industry) introducing timelines for various levels of automation and forecasted market trends for various ADS features for the next 10 years, and the underlying assumptions. On the basis of this analysis, the framework proposes realistic estimates for the envisioned market penetration. Then, from these estimates, the framework suggests a set of realistic scenarios for the different imminent ADS features and possible futures. Operational Design Domains Identifying the ODD of ADS features helps provide a better understanding of how ADS fea­ tures interact with and rely on various infrastructure elements. While technology companies continue to research and develop ADS technologies, IOOs might consider activities that aid the deployment of such technologies. For example, IOOs may look to modify or update infra­ structure design standards and/or policies, maintenance and operations practices, and planning processes. As IOOs make investment decisions that affect these activities, an understanding of the most likely ODDs and their impact on infrastructure will help them prioritize among loca­ tions, infrastructure elements, and roadway types. The identified ADS features cover a broad range of ODDs (e.g., speed range, roadway type, weather, urban roads, and managed lanes) that are likely to impact physical infrastructure. Since these ADS features are likely, in the short to medium term, a realistic prediction, ODDs are key factors in developing realistic deployment scenarios. If multiple ADS features are evaluated at once, it is necessary to identify the ODD of each ADS feature and determine if any ODD elements overlap. Clearly defining the ODD for each ADS feature can help determine appropriate data sources and evaluation methods that consider the combined impacts of all selected ADS features. Specific data and corresponding standards may be needed for IOOs to consider when assessing scenarios. Table 5 suggests a notional approach to breaking down essential metrics IOOs need when evaluating deployment scenarios. Risk Assessment There are a variety of ADS technologies, and each varies in where and how it impacts safety. For example, a low­speed shuttle may positively impact pedestrian safety in a central business district, whereas an SAE Level 3 traffic jam assist feature operating in a dedicated lane will impact safety on arterials and highways. We need to understand the technical and human factors influencing safety. ADSs are projected to decrease incident likelihood and severity but will also change crash populations as ADS behaves differently than humans do. For example, testing and

20 Framework for Assessing Potential Safety Impacts of Automated Driving Systems early deployments of ADS have revealed context­dependent challenges with perception systems in various environmental conditions (e.g., weather) and infrastructure conditions (e.g., lane marking reflectivity) that significantly impact performance. These technologies are associated with various risks and opportunities based on the way they interact with travel behavior, legacy vehicles and infrastructure, and the pertaining human factors. To identify risks, one can consider the limitations of the technology. The following section presents a list of potential risks for Level 3 traffic jam assist (see Table 6). Chapter 4 presents potential risks for five specific ADS features. In addition, to identify opportunities, one can consider the crash types and road­user populations that the technology could impact. Refer to “Map Crash Data to ADS Functionality” in Chapter 2 for further discussion of the potential impacts of ADS by technology and crash type. Example Deployment Scenario The deployment scenario is defined by combining information on the penetration rate, ODD, and limitations of the ADS. Table 6 summarizes elements of an ADS scenario as an example. Step 4—Defining Goals and Hypotheses This section explains how to define goals and hypotheses at two levels: (1) the ADS level and (2) the agency level. ADS-Level Goals and Hypotheses At the ADS level, once a deployment scenario is established, the next step is to define the safety goals (i.e., desired outcomes) and hypotheses related to the specific scenario. Key compo­ nents of the goal may include: Objectives: The goal(s) should indicate the primary and secondary objectives of the ADS technology and the relation to safety. As stated previously, not all ADSs will have a primary goal of improving safety. For example, the primary goal of ADS technologies that support large­truck platooning may be to improve operational efficiency. A secondary goal may be Data thresholds for vehicle-to- infrastructure equipment Dedicated short-range communication data bus thresholds V2X communication standards SAE J2945 Local roadways, densely populated urban areas Infrastructure Assessment Data Needed Standard/Policy Supported ADS ODD Covered Accuracy of mapping services Mapping (global positioning system) accuracy Vehicle-to-everything (V2X) mapping standard Densely populated urban area for obstacle detection and avoidance Performance of ADS in predetermined roadway area Perception sensor metrics (radar/lidar/ camera field of view ranges) ISO 26262 Highways and arterials for ADAS features (lane-keeping assist, adaptive cruise control, etc.) Table 5. Notional data and policy standard assessed by ODD for ADS.

Overview of the Framework Elements 21   to improve safety. If safety is not a primary or secondary goal, then the safety goal should be written as “the goal is to not degrade existing safety performance.” Operational design domains: The goal(s) should carry forward information from the deploy­ ment scenario to help define the target facility type(s) and area(s) of influence. Continuing with the previous example, the goal could be written as “reduce crashes on freeways” to clearly establish the ODD and potential area of impact. The same principles apply when assessing multiple ADSs at the same time with the added nuance to keep track of the applicable target No traffic lights or pedestrians may be present within the relevant viewing range of the vehicle’s sensors.2 Risks • Standards and practices for roadway planning, design, and operations: screenings, lane marking contrast, lighting strategies. • Human factors and road-user interactions: driver response to requests to intervene. • Road user behavior, as well as travel behavior: increase in lax or risky road-user behaviors as people become complacent. • Impacts of mixed modes, vehicle fleets, and different levels of autonomy: dedicated lanes or mixed. • Impacts on law enforcement and emergency response: enabling first responders to interact with ADS when securing a scene. • Communication between legacy and ADS-equipped vehicles (of various levels) and the infrastructure: architectures and standards. • Legislative and regulatory issues: required event data recorder data elements. Opportunities • Enabling pedestrians to be more noticeable to onboard sensors. • Education (e.g., for operators, police, vehicle dealers): new training programs and materials about technology and operational features. • Data (e.g., safety data, reporting, key performance indicators needed): post-crash investigations. • Digital infrastructure (e.g., cybersecurity, mapping, redundancy): work zone data exchange, information sharing and analysis centers. 1https://static.seekingalpha.com/uploads/sa_presentations/758/28758/original.pdf. 2https://www.audi-mediacenter.com/en/technology-lexicon-7180/driver-assistance-systems-7184. Scenario Category Description ADS feature(s) Level 3 traffic jam assist Deployment timeline and market penetration Commercial availability of ADS features (e.g., Audi A8 Level 3 traffic jam assist available in 2018 in Europe), market penetration rate of Level 3 traffic jam assist to increase at 16% compound annual growth rate 2017–2025 according to a 2016 Roland Berger study.1 Deployment context (ODD) The vehicle is on a highway or a multi-lane road with barrier between oncoming lanes and a structure along the edge like guard rails. Slow-moving, nose-to-tail traffic predominates in all neighboring lanes. The vehicle’s own speed must not exceed 60 km/h (37.3 mph). Table 6. Example deployment scenario categories and descriptions.

22 Framework for Assessing Potential Safety Impacts of Automated Driving Systems facility type(s) and area(s) of influence by the ADS. This will support a more detailed analysis in later steps. Target crashes: The goal(s) should indicate the specific safety performance measure to track progress. Safety performance is quantified by the expected number and severity of crashes, potentially by crash type. It could even be as specific as mitigating one or more precipitating events within a sequence of events that could lead to a crash. For example, the previous goal could be more specific by focusing on long­haul, truck­related crashes that are related to driver fatigue or distraction. When assessing multiple ADSs at the same time, the goals could be combined for the overall scenario or kept separate by ADS and ODD. Quantifiable achievement: When possible, the goal(s) should include some measure or threshold to define success. Otherwise, it is difficult to know if and when success is achieved. Continuing with the example, the goal could establish a desired number or percent reduc­ tion for these crashes. If it is difficult to assign a value to the desired change in crashes (e.g., when actual or expected market penetration is unknown), then it will be useful to explore a range of scenarios through sensitivity analysis. Expected timeline: Finally, the goal(s) should include a timeframe for achieving success. This is heavily dependent on the expected timeline for deployment and could include multiple timeframes, depending on the certainty in deployment and penetration rates. With an overall goal in mind, it will be possible to develop hypotheses to test in support of the goal. The hypotheses should help to uncover potential differential impacts on safety. For example, some ADS technologies could mitigate certain crash types and severities while increasing the risk of others. This is particularly important when assessing multiple ADSs simultaneously in the same ODD. A series of questions may be appropriate to help evaluate the hypotheses, especially in cases when the expected outcomes are less clear. This prevents the study from oversimplifying and categorizing results in totality as all good or all bad. For example, the goal of a low­speed shuttle pilot may include improving safety and mobility. The hypothesis may be stated as “The rate of vehicle collisions among travelers going to and from transit station X during the pilot is less than before.” It is also important to ask related questions like “How did the injury and fatality rates from vehicle collisions change?” and “Did the rate of pedestrian and bicyclist collisions change, and to what extent did pedestrian and bicyclist injury and fatality rates change?” The series of questions could also include questions related to human behavior and human factors as many of the related safety relationships have not been quantified or are not well understood. Related questions may include “How does driver performance change over time as the proportion of ADS­operated vehicle­miles increases?” and “How do road­user behaviors change (e.g., increase in lax or risky road­user behavior) as the proportion of ADS­ operated vehicle­miles increases?” A crash sequencing exercise is also useful to think through the contributing factors and pre­ cipitating events that lead to a crash. This can help to identify the truly correctable crashes related to the ADS scenario of interest. Similar to the “before crash” row of the Haddon Matrix, mapping the crash sequence of events is an exercise of thinking through the timeline of events and factors that lead to a crash. The difference is that crash sequencing is focused on one crash or a specific type of crash. In the context of this framework, it is helpful to think of the potential interaction of the ADS feature(s). For each event in the sequence, ask if the ADS feature(s) could address the event and thereby mitigate the crash. The following are a few examples of crash sequences. Example 1: sequence of events for vehicle-pedestrian crash at intersection: Pedestrian enters crosswalk against the pedestrian signal Driver in vehicle 1 is distracted and does not see the pedestrian Crash between vehicle 1 and pedestrian

Overview of the Framework Elements 23   Example 2: sequence of events for vehicle-vehicle crash at intersection: Mapping the crash sequence is also useful for identifying potential increases in collisions with the introduction of ADS. The following is a continuation of the pedestrian crash example above but now considers an ADS scenario where vehicle 1 is equipped with forward collision warning but vehicle 2 is not. In this case, vehicle 1 is able to stop in time to avoid the collision with the pedestrian, but this creates another sequence in the chain of events as vehicle 2 is not able to stop, resulting in a rear­end collision. Example 3: sequence of events for ADS-vehicle crash at intersection: Agency-Level Goals and Hypotheses Agencies set many goals that can range from high­level strategic goals down to highly spe­ cific and localized performance targets. These goals help capture transportation user needs and agency priorities that the agency hopes to achieve through specific actions or countermeasures (e.g., deployment of an ADS feature). At the agency level, all states have developed a Strate­ gic Highway Safety Plan (SHSP). SHSPs establish statewide safety goals and identify specific emphasis areas and strategies that should combine to achieve this overarching goal. For example, Washington State established a goal of zero deaths and serious injuries. Washington also includes a detailed discussion of new technology and traffic safety in the SHSP. The plan recognizes tech­ nological advancements but does not quantify the potential impact of this technology on safety. SHSPs can be used as a foundation or starting point to identify target safety goals. These plans can also help to identify the priority safety issues (emphasis areas) and indicate whether the state has identified ADS or related technologies as a strategy to address one or more emphasis areas. At the agency level, this framework can be adapted to test hypotheses related to the effec­ tiveness of individual ADS technologies, the impacts of infrastructure and operations on ADS safety performance, and the potential contribution of ADS to statewide goals. This can help agencies to understand the potential contributions of ADS toward meeting the state’s overall safety goals or the specific safety goals within an emphasis area. Given that the goals are already established in the SHSP, the application of this framework becomes an exercise of defining and testing hypotheses. For example, the Washington State SHSP indicates that “the anticipated benefits of [autonomous] vehicles include decreased crashes, increased mobility, and an increase in fuel efficiency.” The hypotheses to test are numerous, but one opportunity to focus the hypothesis testing is to develop hypotheses related to established emphasis areas. For example, Table 7 presents hypotheses related to common emphasis areas included in state SHSPs. While these hypotheses incorporate crash­based performance measures, it is also possible to develop hypotheses related to other performance metrics, such as those related to public health, human behavior, and human factors. For example, the last row in Table 7 relates to aggressive driving crashes, but one or more additional hypotheses could be developed to explore Vehicle 1 stops in left-turn lane at signalized intersection Vehicle 2 stops in opposing left-turn lane, obstructing the view of driver in vehicle 1 Vehicle 3 is traveling through the intersection, approaching from same direction as vehicle 2 Driver of vehicle 1 turns in front of vehicle 3 Crash between vehicle 1 and vehicle 3 Pedestrian enters crosswalk against the pedestrian signal Driver in vehicle 1 is distracted and does not see the pedestrian ADS feature activates to stop vehicle 1 Vehicle 2 is trailing vehicle 1 and driver does not react in time Crash between vehicle 1 and vehicle 2

24 Framework for Assessing Potential Safety Impacts of Automated Driving Systems aggressive driving behaviors, which are a precursor to the crashes (e.g., will conditional traffic jam assist reduce driver frustration and erratic lane­change maneuvers?). Finally, it is important to identify potential processes and procedures that could impact or be impacted by the goal in the presence of ADS. Table 8 provides example questions to help answer hypotheses for how changes in planning, design, and operations processes and procedures will impact safety goals. It is helpful to recognize that capabilities can vary across state and local agencies, and questions may accommodate the various levels of maturity. While SHSPs are a good starting point for identifying safety goals and developing hypotheses, it is important to consider other sources of safety goals, as shown in Table 8, since SHSPs are typically updated only every 5 years. Step 5—Choose Method of Analysis Among all benefit types, traffic safety is envisioned to be the major benefit of ADS technologies. Accordingly, the following subsections discuss the approach for estimating the safety benefits from an envisioned ADS feature. Traffic Safety Impacts Information about the benefit analysis for traffic safety savings is expressed in monetary value. This approach requires an estimate of the number of crashes avoided due to the adoption of ADS. The magnitude in traffic safety savings corresponds with the market penetration of ADS to the larger network. For instance, if ADS­equipped trucks represent 10% of the truck vehicle fleet, then it might be assumed that ADS­equipped trucks could mitigate approximately 10% of truck­related crashes. This assumes the ADS technologies work in 100% of crashes, which may not be a reasonable assumption. Thinking back to earlier steps, it is useful to narrow the population of crashes to “correctable crashes.” This would include only those crashes within the ODD and those where the ADS feature(s) of interest could address specific events within the sequence of events that led to the crash. The KABCO injury scale also can be used for establishing crash costs. This scale was devel­ oped by the National Safety Council (NSC) and is frequently used by law enforcement for classifying injuries: • K—Fatal, • A—Incapacitating injury, • B—Non­incapacitating injury, • C—Possible injury, and • O—No injury. Emphasis Area Hypothesis Lane departure Will conditional automated highway drive reduce lane departure crashes? Intersections Will conditional traffic jam assist drive increase rear-end crashes between ADS- equipped and non-ADS-equipped vehicles? Pedestrians Will conditional automated highway drive reduce pedestrian crashes in urban areas? Work zones Will conditional automated highway drive increase work zone crashes where signing and markings are not clearly identifiable? Aggressive driving Will truck platooning increase aggressive driving crashes on freeways? Table 7. Specific hypotheses related to common SHSP emphasis areas.

Overview of the Framework Elements 25   Dimension Example Processes and Procedures Example Questions Planning • Strategic Highway Safety Plan (SHSP). • Regional intelligent transportation system (ITS) architecture. • ITS Strategic Plan. • Transit Safety Plan. • Nonmotorized (bicycle and pedestrian) plan. • Metropolitan transportation plan statewide and regional long-range transportation plan. • Public involvement plan and public participation plan. • Freight plans. • Financing plans. • Performance management targets. • What types of trips and geographic areas will early ADS deployments favor? • How do we coordinate across the agency, state, and nation? • How do we balance capabilities in operations of ADS with safety? • What incremental strategies can we invest in now so that ADS will continue to provide benefits as the market grows? • What ADS market penetration do we need to evaluate and analyze operational benefits? • How do we manage expectations and not oversell capabilities? Design • Standards and best practices for infrastructure markings, signage, geometry, countermeasures, human factors (bike and pedestrian). • Technology choices, (e.g., materials, interoperability). • Vehicle-to-infrastructure equipment and standards. • Training, knowledge, information sharing. • Communications security and privacy policy. • What are the safety impacts of design decisions (e.g., what is the difference in the expected safety performance of two viable designs that meet design standards)? • What are the safety impacts and mobility tradeoffs of different intersection traffic control alternatives (e.g., four-way stop- control, two-way stop-control, traffic signal, roundabout)? • What are the safety impacts of design variances and exceptions (e.g., what is the change in safety if a design element does not meet the related design standard)? • What are the safety impacts of implementing safety countermeasures to enhance safety performance at a given location? • Where can design standards be relaxed? • What skills and training are needed for staff, and what is the best delivery format for training? Operations • Traffic control device operation. • Access point management. • On-road ADS testing. • Emergency response procedures. • Law enforcement procedures. • Security processes and procedures. • How will vehicles respond to dynamic conditions, (e.g., weather, work zones, and roadway lighting)? • What are the safety impacts and mobility tradeoffs of different traffic signal phasing alternatives (e.g., permitted versus protected left-turn phase)? • Training, knowledge, information sharing. • How to justify decisions to control or consolidate access points? • How do we leverage opportunities for multimodal benefit—including transit, traditional vehicles, pedestrians, bicyclists, fleets? • What do emergency medical services and law enforcement need to know to control and prepare for an emergency scene? • Will new types of risks be introduced such as limitations in technology, cybersecurity breaches, overreliance on technology that could affect the operations? • How do we embrace interoperability? Table 8. Example questions to test hypothesis.

26 Framework for Assessing Potential Safety Impacts of Automated Driving Systems There are several sources of crash costs by severity level, including the Highway Safety Manual (HSM) and FHWA’s Crash Costs for Highway Safety Analysis (Harmon et  al., 2018). Both resources provide crash costs based on the KABCO scale, but the values in FHWA’s Crash Costs for Highway Safety Analysis are based on more recent data and are shown in Table 9. If a state has not developed its own crash costs, these costs could be used to calculate safety benefits. Analysts can use crash costs to convert expected ADS safety benefits into monetary terms as demonstrated in later sections. The sections below describe how to identify appropriate data sources, define metrics and evaluation criteria, and select an appropriate evaluation method to test the hypotheses developed in the previous step. Define Data Sources Several pilots are under way across the United States to explore the feasibility and challenges associated with different ADS applications. Still, there are limited ADS datasets that can directly help agencies understand expected safety performance under different ADS scenarios. With limited ADS safety data available, four data categories can prove useful: (1) traditional crash datasets, (2) traditional roadway datasets, (3) advanced safety datasets, and (4) ADS datasets. Traditional Crash Datasets Traditional crash datasets provide information on reported crashes, including the date and location of the crash, the type and severity of the crash, and crash contributing factors based on the reporter’s investigation of the event. The crash contributing factors may include information on driver and other road­user behaviors, vehicle type and characteristics, roadway attributes, and environmental conditions. The specific variables, amount of detail, and level of accuracy vary by dataset. The following are some of the primary opportunities to use traditional datasets in assessing the safety impacts of ADS. 1. Establish baseline safety performance for traditional vehicle fleet: In scenario planning and alternatives analysis, the future performance for the do­nothing scenario is estimated. In this case, traditional crash datasets help quantify and assess the safety performance of traditional vehicles and identify the conditions under which crashes are occurring. 2. Estimate change in number, type, and severity of crashes under different ADS scenarios: Traditional crash datasets include information on the number, type, and severity of crashes. If the relationships between the presence of a given ADS technology and the probability of Severity Level Crash Cost Fatality (K) $15,134,325 Disabling injury (A) $851,749 Evident injury (B) $282,550 Possible injury (C) $163,792 Property damage only (O) $16,175 1https://safety.fhwa.dot.gov/hsip/resources/fhwasa09029/sec4.cfm. Table 9. Crash cost by injury severity level (FHWA’s Crash Cost for Highway Safety Analysis).1

Overview of the Framework Elements 27   different crash types are known (or assumed), then these datasets can be used to quantify the number of crashes by type and severity that could be impacted by the various ADS technologies under different deployment scenarios. 3. Understand crash contributing factors for crashes involving ADS-equipped vehicles: Traditional crash datasets typically provide some level of detail on the vehicles involved, including the make, model, and year. Knowing the standard ADS features on specific vehicle types, the traditional safety datasets can be used to search for specific vehicle types and assess the factors involved in the crash. This could help identify factors that are common among crashes involving specific technologies. For example, are there common roadway or envi­ ronmental conditions that are associated with the involvement of vehicles with certain ADS technologies? One limitation is that these datasets will not likely indicate if a vehicle has an aftermarket system or if the driver had turned off the ADS technology (if that is an option for the given technology). Agencies may want to investigate human factors while analyzing crash contributing factors. These are factors that contribute to a crash and are directly attributable to the driver, such as inattention, distraction, fatigue, and impairment from drugs or alcohol. Human factors that apply to traditional vehicles may not apply to ADS; however, other human factor challenges may arise with greater ADS deployment. These human factors may include drivers’ ability to take back control from ADS, legacy vehicle drivers’ interaction with ADS­equipped vehicles, and other road users’ ability to identify and interact with ADS. Common datasets under this category include Fatality Analysis Reporting System (FARS), General Estimates System, Highway Safety Information System (HSIS) crash data, and state and local crash datasets. For example, if the hypothesis is focused on fatal crashes, then the national FARS or the state crash database may be appropriate sources of data. If there is a need to focus on specific facility types and determine the roadway characteristics or average weather conditions, then there will be a need to merge data from other sources such as the state roadway inventory or the FHWA Weather Data Environment. Traditional Roadway Datasets Traditional crash data can be augmented with data describing the roadway characteristics at the site of a crash or across a given area or region. Numerous factors contribute to or are associated with the frequency of crashes, including behavioral, vehicular, and roadway charac­ teristics. In many cases, a crash can result from several factors. Joining roadway datasets with crash datasets can maximize the coverage of these factors and provide a clearer picture of safety performance. There are several potential applications of roadway datasets in estimating the safety impacts of ADS. Often, ADS technologies rely on camera sensors to detect roadway features such as lane striping or crosswalks to keep the vehicle in the correct position on the road. Rich roadway datasets could allow ADS companies and agencies to screen roadways and determine locations that would be candidates for ADS operation, or roadways that might not be desirable for ADS operation. Roadway datasets can also help to identify the population of current crashes that could be impacted. For example, by joining the crash and roadway data, it is possible to identify the number of crashes by crash type that are occurring on different roadways and at different roadway site types (e.g., intersections, horizontal curves, vertical curves). A generic assumption related to ADS might be that a technology will address all crashes of a certain type. However, if the roadway conditions do not allow that technology to operate properly, it is necessary to

28 Framework for Assessing Potential Safety Impacts of Automated Driving Systems disaggregate ADS safety performance estimates with and without those conditions. Datasets under this category include Highway Performance Monitoring System, HSIS roadway data, and state and local roadway inventory files. Advanced Safety Datasets Advanced safety datasets, created for safety research, include Strategic Highway Research Program (SHRP2) Naturalistic Driving Study (NDS) dataset, SHRP2 Roadway Information Database, Motorcycle Crash Causation Study, Large Truck Crash Causation Study, and National Motor Vehicle Crash Causation Study. ADS Datasets The growing presence of ADS technologies and companies has led to the proliferation of data­ sets related to ADS testing and operation. Many of these datasets originated with companies working to develop ADS technology and train ADS in object identification and other skills. A large subset of these datasets is intended for use in training and benchmarking (i.e., conducting extensive testing to develop a standard for state of the practice). These datasets may provide insights into not only ADS safety performance with the DDT, but also interactions with humans. In addition to these training and benchmarking datasets, several public agencies have begun to implement data systems for addressing the growing ADS environment. They include Camera­ Based Self­Driving Object­Detection Datasets, KITTI 3D Object Detection Dataset, nuScenes, Specialized Computer Training Benchmark Datasets, Waymo Open Dataset, Lyft Level 5 Dataset, California DMV Autonomous Vehicle Data, Colorado DOT Autonomous Vehicle Data, and Warrigal Dataset. Define Metrics It is essential to define metrics that appropriately capture the ideas presented in the hypotheses and questions. It is also important to note that some metrics, while theoretically computable, may not actually be producible with available data. Researchers are limited by data granularity and availability and may not be able to calculate all desired metrics. Analysts should start by defining preferred metrics and then assess the feasibility of computing those metrics. This is an important step because this process can reveal data gaps that can guide future data collection efforts and potentially be addressed in further research. The metrics relate to the safety components that are the output of the analysis. Metrics may include safety outcomes, such as crashes, or predictive indicators, such as average time­to­ collision violations at an intersection. Crash­based metrics can include a change in crash fre­ quency and/or severity. If specific crash types are anticipated to be affected, they should also be stated as a part of the analysis metrics. The method for identifying the crashes or events of interest is important to state and describe. For example, the crashes of interest might be identified through categorizing and filtering the crashes based on the ODD and specific assumptions. The frequency of crashes can be measured based on occurrences normalized by an exposure metric, for example, crashes per pedestrian encounter. Crash types or situations that are exceedingly rare to experience, and therefore potentially challenging to identify or quantify, are commonly referred to as “edge cases.” Other types of safety outcome metrics include public health indi­ cators, such as travel impact assessment. For example, trip rate for persons with disabilities and economic value of those trips can serve as an indicator of societal impact from ADS deployment (Baker et al., 2017). When assessing multiple ADS features at once, it is necessary to identify target crashes related to each ADS feature and identify any overlap between the target crash types within a given ODD.

Overview of the Framework Elements 29   This avoids double­counting crashes in the analysis. Potential safety benefits are not simply addi­ tive between the ADS features. Whether the ADS features are redundant, complementary, or contradictory should be considered If the ADS features are redundant (i.e., address the same crash type within the same ODD), then it is appropriate to consider the safety benefits of one of the ADS features (presumably the one with the greatest benefits). If the ADS features are comple­ mentary (i.e., address different crash types or the same crash types but in different ODDs), then it is appropriate to consider the safety benefits of both ADS features, ignoring any redundant effects. If the ADS features are contradictory (i.e., impact the same crash types in the same ODD but in opposite directions), then it is appropriate to consider the safety impacts of both ADS features; however, in this case, the impacts would offset to some extent. Figure 7 illustrates these scenarios. Metrics related to human behavior and human factors also relate to safety and could be considered when assessing the safety performance of ADS features. While ADS features present a tremendous opportunity to improve safety, there is also the potential for challenges related to human factors and limitations of the technology. For example, ADS technology may not operate well in all road and weather conditions. If the human operator does not understand the limitations of the technology, there is the potential for an increased delay in responding to requests for manual takeover of vehicle control. Conversely, if the human operator understands the limitations of the technology (e.g., does not operate in rain or snow), then the human opera­ tor may be more alert and ready to take control if those conditions are expected along the route. Cunningham and Regan (2015) indicate that drivers must understand the capabilities (and limita­ tions) of ADS technologies and must maintain situational awareness to improve the chance of safe operation. They also discuss the following factors related to ADS technologies, which could be used to define related metrics: Driver inattention and distraction: Driver inattention and distraction can pose issues when using ADS features. When cognitive workload is too low, which can occur when ADS features are engaged, drivers can experience passive fatigue (Desmond and Hancock, 2001), and drivers might engage in other tasks. This creates safety­related concerns for scenarios when a driver needs to take (or receive) control of the vehicle (Cunningham and Regan, 2015). Situational awareness: When ADS features are engaged, drivers might pay attention some­ where else and engage in other tasks. In these situations, drivers may not be fully aware of the environment around them, including the vehicle and road (De Winter et al., 2014), which can lead to issues if the driver needs to take (or receive) control of the vehicle. Overreliance and trust: When using ADS features, drivers might become over reliant on the features. Drivers may rely on the technology (Cunningham and Regan, 2015), but if the reliance is too high, drivers might assume the feature will alert them if something is wrong or if they need to take control when it might not (Parasuraman and Riley, 1997). Figure 7. Examples of redundant, complementary, or contradictory ADS effects (Source: VHB).

30 Framework for Assessing Potential Safety Impacts of Automated Driving Systems Skill degradation: When drivers use ADS features frequently and do not use their own driving skills often, they might lose or diminish their driving skills (Parasuraman et al., 2000). Motion sickness: The use of ADS features could also cause motion sickness to the driver or occupants of the vehicle. This is due to drivers losing control over the actions of the vehicle (Rolnick and Lubow, 1991) and losing awareness of what actions will be made next (Golding and Gresty, 2005). Evaluation Method The defined metrics are used with an evaluation method to estimate the safety impact of each ADS feature. The evaluation method needs to be stated and described. For example, if the evaluation method is a percent crash reduction using total crashes and the specific crashes expected to be impacted by an ADS feature, an equation for the percent crash reduction should be included in the evaluation method. The evaluation method can also be further disaggregated by specific crash type and severity levels. In addition, artificial intelligence/machine learning and big data solutions can help evaluate and parse large volumes of safety­related data (e.g., probe data, NDS data) efficiently. The HSM is one opportunity to support the evaluation method. Specifically, Part C of the HSM provides a predictive method for several facility types. The predictive method allows analysts to estimate the number of crashes for alternative scenarios based on roadway and traffic characteristics. This is useful for establishing the safety performance for a base condition (e.g., existing conditions or future conditions with changes to the infrastructure or traffic volume). The analyst could use the predicted crashes for the base condition in conjunction with other methods to estimate the expected change in crashes assuming different ADS scenarios and penetration rates. IHSDM is an advanced safety analysis tool that uses the HSM Part C predictive method to estimate the safety performance of existing and future conditions under the current vehicle fleet. Users can input roads and road networks into IHSDM to automatically analyze their safety per­ formance. Figure 8 displays an example of a road network input into IHSDM, which includes road segments and intersections. Different road elements (e.g., roadway geometry or traffic conditions) can easily and quickly be adjusted (using the editor in Figure 9) to obtain new outputs and compare safety performance results. Users can change road and traffic characteristics to account for ADS features to obtain their potential safety impact prior to deploying an ADS feature. While the predictive method in the HSM and associated safety analysis tools provide an opportunity to estimate the safety performance of specific roadway design and traffic operations scenarios, the methods were developed without capturing the impacts of ADS features. To dif­ ferentiate the use of this framework, the primary use is for planning­level decisions that are more likely associated with Part B of the HSM. The evaluation method presented in this framework is not meant to replace the HSM predictive method, but over time, could provide valuable insights for updating or adjusting the HSM predictive method. For example, by tracking ADS implementations and evaluating the post implementation effectiveness of ADS, it might be possible to produce adjustment factors (or CMFs) that represent a change in crashes due to certain ADS features. The result from this framework could also be used in conjunction with the predictive method in the HSM to compare the safety performance associated with the different scenarios, including scenarios with different levels of ADS deployment. The monetary value of ADS safety benefits would be the difference in estimated crash costs between the scenario with a specific ADS feature deployed versus the estimated crash costs without the ADS feature. The following equation can be used to identify the monetary value of

Overview of the Framework Elements 31   Figure 8. Example of road network in IHSDM (Source: IHSDM Project Output).

32 Framework for Assessing Potential Safety Impacts of Automated Driving Systems crashes, for a particular ADS feature, as a function of crash severity type, ADS market penetra­ tion, corresponding crash cost, and CMF for ADS. ∑=Monetary value of crashes after p p pADS deployment N MP Cost CMFCrash ADS sev ADSsev Eq.1 where: NCrash = Total crash by severity type (crash count by type) MPADS = ADS market penetration (decimal) Costsev = Monetary cost by severity type (dollar value) CMFADS = Crash modification factor for ADS implementation (decimal) CMF is defined in the HSM as a value that quantifies the expected change in crash frequency at a site as a result of implementing a specific countermeasure. Countermeasures can also be called “treatments” or “safety treatments.” A CMF can estimate the expected change in crash frequency for total crashes, a particular crash type, or a particular severity. CMF is expressed as the following: =CMF Expected crash frequency with treatment Expected crash frequency without treatment Eq. 2 Currently, there are no CMFs for ADS features; however, as demonstrated later in the frame­ work (Chapter 4), state and local officials can apply their understanding of the ADS technology to historical crash data to estimate the number and types of crashes that could be mitigated by ADS. From these values, agencies could estimate a CMF for ADS. Figure 9. Example of IHSDM Highway Editor used to change road characteristics (Source: IHSDM Project Output).

Overview of the Framework Elements 33   Historical crash data broken down by severity type following the KABCO classification can be obtained from local and state law enforcement or transportation agencies. The data should be classified by annual occurrence to find the annual vehicle crash cost and savings. The crash savings can be expressed as the following equation: = −ADS safety benefits Crash cost without ADS deployment Crash cost with ADS deployment Eq. 3 For more information on converting crash­based estimates to monetary benefits, refer to FHWA’s Highway Safety Benefit-Cost Analysis (Lawrence et al., 2018). Results The previously defined evaluation methods are used with the crash data to estimate the safety impact of the various ADS features. The results can be disaggregated by timeline (e.g., separate analysis for short­ and medium­term timelines). Typically, the different timelines have different ODD elements and different VMT share. The results should highlight this difference, which also shows how the expected safety performance changes based on the timelines and if the ODD is expanded and VMT share changes. A sensitivity analysis can be included in the analysis to estimate the potential range of safety impacts based on varying penetration levels, expected ADS effectiveness (and failure rates), and other assumptions. For instance, the sensitivity analysis can include different assumptions to estimate the safety impact of ADS features if the VMT share is not fully reached or to estimate the impact at different penetration levels. Similarly, the analyst can adjust the expected level of effectiveness (e.g., percent change in crashes) and potential failure rate of ADS to determine how these assumptions affect the results under different deployment scenarios. For example, it might be of interest to estimate the change in the number of crashes by severity assuming 100% effec­ tiveness, 75% effectiveness, and 50% effectiveness. When the deployment scenario includes multiple ADS features, and these ADS features provide redundancy in terms of the target crashes, it may be appropriate to assume a higher level of confidence and higher percent effectiveness in the sensi­ tivity analysis. Further, it may be appropriate to assume a higher percent effectiveness as the analysis focuses on specific crashes under specific conditions (i.e., those most likely corrected or mitigated by the particular ADS application). The sensitivity analysis can also help account for human behaviors and human factors (or related uncertainties) and the expected change in safety performance. Specific examples of human behavior and human factor considerations are discussed with respect to each ADS feature in Chapter 4. The following paragraphs present an example analysis of potential fatalities mitigated with the implementation of low­speed ADS shuttles, at the national level, with mitigating fatalities. The US DOT maintains detailed fatality reports through the NHTSA FARS database. Table 10 shows a breakdown of the reported vehicle fatalities for year 2018 reported to the NHTSA FARS database. The implementation of a low­speed ADS shuttle is expected to reduce bus­related vehicle crashes. For the following example, it is assumed the implementation of the ADS shuttle will reduce the risk of all types of bus­related vehicle crashes by 70%. This value should be updated as more research is completed on the safety impacts of ADS­equipped buses. A 70% reduction in the risk of vehicle crashes is equivalent to a CMF value of 0.3. Figure 10 and Table 11 show the monetary costs and savings in bus fatalities for various ADS market penetrations, assuming a growth factor of 2%. The baseline of 234 bus­related fatalities in Table 10 is used to substitute the Ncrash as the total annual fatality­related crashes described in Equation 1. The same methodology described in the examples shown above that analyzes the cost and savings at a national scale over the market penetration period of ADS can be scaled down to a

34 Framework for Assessing Potential Safety Impacts of Automated Driving Systems Body Type Year 2018 Passenger cars 20,333 Light trucks 19,775 Large trucks 4,862 Motorcycles 5,115 Buses 234 Other/unknown 1,553 1https://www-fars.nhtsa.dot.gov/Vehicles/VehiclesAllVehicles.aspx. Table 10. National fatalities by vehicle type for year 2018 (Fatality Analysis Reporting System).1 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% $0 $50 $100 $150 $200 $250 Y ea r 1 Y ea r 3 Y ea r 5 Y ea r 7 Y ea r 9 Y ea r 1 1 Y ea r 1 3 Y ea r 1 5 Y ea r 1 7 Y ea r 1 9 Y ea r 2 1 Y ea r 2 3 Y ea r 2 5 Y ea r 2 7 Y ea r 2 9 Y ea r 3 1 Y ea r 3 3 Y ea r 3 5 Y ea r 3 7 Y ea r 3 9 M ar ke t P en et ra tio n Fa ta lit y C os t M ill io ns Total Bus Cost without AV Total Bus Cost with AV Total Bus Savings with AV Deployment Figure 10. National monetary cost due to bus-related fatalities and savings due to crash reduction (Source: Booz Allen). Bus Fatality Costs Market Penetration 0% 20% 40% 60% 80% 100% Bus fatality cost without ADS $3,541 $3,988 $4,149 $4,316 $4,766 $7,368 Bus fatality cost with ADS $3,541 $3,450 $2,856 $2,479 $2,063 $2,210 Bus fatality savings with ADS $0 $538 $1,290 $1,837 $2,702 $5,158 Table 11. National monetary costs and savings (millions) in bus fatalities by ADS market penetration.

Overview of the Framework Elements 35   defined regional scale. The approach was applied to the state of Minnesota; Table 12 lists the Minnesota fatalities by vehicle type for the year 2018. The example scenario in Figure 11 shows the monetary costs and savings due to bus­related fatalities and savings, respectively, over 40 years of ADS deployment to achieve 100% market penetration for the state of Minnesota, assuming a 2% growth rate. Table 13 shows the monetary costs and savings in bus fatalities by various ADS market penetra­ tions for the state of Minnesota. The baseline of seven bus­related fatalities in Table 12 is used to substitute the NCrash as the total annual fatality­related crashes described in Equation 1. The example above demonstrates that by using the same framework applied nationally, a study can be tailored for a specific geographic region to assess local impacts of monetary savings due to ADS deployment. The approach demonstrates that scope can be changed simply by using available local historical values corresponding to the NCrash. The NCrash and CMFsev can also be adjusted to assess the safety impacts of ADS deployment on subsets of vehicle collision types (e.g., rear­end, sideswipe, etc.). This detailed breakdown should be applied to improve crash analysis primarily when CMF values provide a breakdown at the collision type level. Body Type Year 2018 Passenger cars 605 Light trucks 595 Large trucks 105 Motorcycles 143 Buses 7 Other/unknown 25 Table 12. Minnesota fatalities by vehicle type for year 2018 (Fatality Analysis Reporting System). 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% $0 $50 $100 $150 $200 $250 Y ea r 1 Y ea r 3 Y ea r 5 Y ea r 7 Y ea r 9 Y ea r 1 1 Y ea r 1 3 Y ea r 1 5 Y ea r 1 7 Y ea r 1 9 Y ea r 2 1 Y ea r 2 3 Y ea r 2 5 Y ea r 2 7 Y ea r 2 9 Y ea r 3 1 Y ea r 3 3 Y ea r 3 5 Y ea r 3 7 Y ea r 3 9 M ar ke t P en et ra tio n Fa ta lit y C os t M ill io ns Total Bus Cost without AV Total Bus Cost with AV Total Bus Savings with AV Deployment Figure 11. Minnesota monetary cost due to bus-related fatalities and savings due to crash reduction (Source: Booz Allen).

36 Framework for Assessing Potential Safety Impacts of Automated Driving Systems Bus Fatality Costs Market Penetration 0% 20% 40% 60% 80% 100% Bus fatality cost without ADS $105 $119 $124 $129 $142 $220 Bus fatality cost with ADS $105 $103 $85 $74 $62 $66 Bus fatality savings with ADS $0 $16 $38 $55 $81 $154 Table 13. Minnesota monetary costs and savings (millions) in bus fatalities by ADS market penetration. The expected safety performance results can be used with estimated investment costs to perform a benefit­cost analysis. As illustrated in the example application of this framework in Chapter 4 with ADS­equipped trucks (Levels 3 and 4), this can help to determine economic impacts of investment decisions or to compare various alternatives (e.g., compare different investment decisions). In a safety­focused benefit­cost analysis, the benefits, such as a decrease in crashes, are first converted to a monetary value based on average crash costs. The benefits could also include other factors such as operational and mobility, public health, and envi­ ronmental impacts. The monetary benefits are compared with the costs associated with ADS implementation (e.g., infrastructure costs to enhance ADS performance or to expand the ODD, thereby expanding the benefits). ADS implementation costs could include costs associated with new infrastructure elements, upgrading infrastructure, and ongoing maintenance. For further information, refer to FHWA’s Highway Safety Benefit-Cost Analysis Guide (Lawrence et al., 2018). Further Socioeconomic Considerations While safety benefits are the main focus of this document, ADS technologies, especially the Level 4 and Level 5 applications, have the potential to provide benefits to public health more broadly than just reduction in crashes. For example, these technologies are expected to have considerable socioeconomic impacts (travel, public health, and environmental) beyond just a reduction in crashes. ADS improves access to business districts and commercial and medical facilities for the elderly and people with disabilities. Once ADS technologies are deployed, induced demand would be served by these technologies, and induced travel would translate as more travel trips showing on the network. Induced demand is the demand that has not been previously accommodated by the network due to capacity, convenience, cost, safety, disability, or age constraints. IOOs should be aware of the landscape of the socioeconomic factors that are affected by ADS features. Accordingly, this section discusses the approaches that IOOs need to consider for identifying these socioeconomic impacts of ADS. However, the scope of the framework application in Chapter 4 focuses only on the safety benefits. Induced Travel Trips To calculate the socioeconomic impact of ADS technologies, IOOs need to estimate the induced travel trips triggered by ADS deployment. The induced travel trips may be considered as the degree to which mobility is improved and driving constraints are removed on the network. This translates into new additional travel trips that could be accommodated. A portion of these trips would be zero­occupancy trips, where the vehicle is empty while moving in the traffic

Overview of the Framework Elements 37   heading to a passenger or even cruising waiting for a passenger. An estimate for the non­zero­ occupancy induced trips could be calculated using the following equations: = × × × Eq. 4Induced travel trips Base trip rate Trip rate impact Market size 365 DaysNon_Zero Equation 4 may need to be calculated for the different population groups: • Persons with disabilities. • Older adults. • All other travelers. The base trip rate is the average number of trips per person per day. The nationwide average base trip rate is 4.06 (US Census Bureau, 2012). Planning departments within state and local agencies may be able to ascertain more accu­ rate estimates of the base trip rate based on local travel behavior, car ownership, and demo­ graphics data. The trip rate impact is the percentage increase in daily trips from the base trip rate. This factor expresses the ability for an intervention (i.e., ADS technology) to overcome challenges, such as accessibility, reliability (e.g., measured in buffer time), and affordability of transportation. A study by the Puget Sound Regional Council (Childress et al., 2014) estimated the trip rate impact of autonomous vehicles for all travelers in the Puget Sound region would be 4.88%. Planning departments within state and local agencies may be able to ascertain more accurate esti­ mates of trip rate impact based on their local travel conditions, challenges, demographics, and ADS market penetration data. The market size is the population within the deployment region that has the potential to benefit from the mobility benefits of the ADS technology. For example, the market size may be estimated as the population within the city limits for a central business district deployment. This estimate may be informed by the deployment geography and targeted demographics. One approach to obtaining more accurate travel impact estimates is to segment the popu­ lation based on how ADS provides mobility benefits. Persons with disabilities may benefit dif­ ferently than a tourist or daily commuter. For example, the base trip rate for the population with disabilities is approximately 2.60 (Mattson, 2012). This population is likely to have a higher trip rate impact factor, evidenced by a separate study by researchers in Carnegie Mellon University that estimated the impact of autonomous vehicles on vehicle miles traveled among non­drivers, elderly drivers, and drivers with travel­restrictive medical conditions would be a 14% increase (Harper et al., 2016). Estimating the zero­occupancy induced trips would be more challenging for IOO and planners. The zero­occupancy trips could be estimated as a percentage of the total non­zero­ occupancy trips. However, this percentage would vary significantly depending on the nature of ADS feature deployment. For instance, if Level 4 ADS features are deployed as privately owned vehicles, the zero­occupancy trips percentage is anticipated to be much higher than that if the ADS features are deployed as shared robo­taxis, operated by ride­hailing companies. Moreover, there are additional complexities to the estimation of the non­zero­occupancy trips that relate to the inexpensive operating costs of an electric ADS­operated vehicle. For instance, it would cost owners of an ADS­equipped vehicle about 50 cents per hour for their empty vehicle to cruise city­center streets at urban traffic speeds (Coxworth, 2019). That is much less expensive than paying to park at a metered area or in a lot. This could encourage privately owned noncommercial ADS­operated vehicles to add a significant number of zero­occupancy trips to the network. Dynamic pricing and congestion fee policy set by the planners are envi­ sioned to play a pivotal role in dampening the percentage of zero­occupancy induced trips.

38 Framework for Assessing Potential Safety Impacts of Automated Driving Systems Economic Value of Trips Estimates of the economic value of each non­zero­occupancy trip can be based on guidance from the Federal Emergency Management Agency, which assumes a delay time of half­day (12 hours) per trip to reflect the loss in productivity and spending for each trip that is not made (i.e., when the trip does not happen, the economy is less productive and there is less spending overall). Therefore, this analysis assumes that the loss of a trip is equal to a trip not taken— each having a similar impact on the overall economy. Since the value of time for local travel for personal and business purposes is $13.85 per hour in 2016 dollars, each new non­zero occupancy trip represents a gross benefit of approximately $166 (US DOT, 2017). The gross economic benefit per trip considers the growth and wealth creation that benefits everyone, not only the traveler (health providers, retail stores, transportation companies, etc.). Economic value may be expressed as a percentage of the gross domestic product. The economic value of zero­occupancy trips is anticipated to comprise solely the vehicle operating costs and the network­induced congestion costs. Dynamic pricing and congestion fee policies envisioned by cities can be used to estimate the economic cost of zero­occupancy trips. No major economic benefits are expected from the zero­occupancy trips. Economic Impact of ADS Due to Induced Demand The economic impacts of ADS can be calculated as the result of subtracting the cost of the induced zero­occupancy trips from the benefits of the induced non­zero­occupancy trips. The benefits can be calculated as the product of the newly induced travel trips served by ADS and the economic value of each trip. Impact of ADS 5 Benefits of induced travel tripsNon-Zero 2 Cost of induced travel tripsZero Benefits of induced travel tripsNon_Zero 5 Induced tripsNon_Zero 3 Economic value of tripsNon-Zero Cost of induced travel tripsZero 5 Induced travel tripsZero 3 Economic value of tripsZero Environmental and Fuel Consumption Considerations. ADS technologies are envisioned to have positive environmental impacts on the transportation network. The main impacts expected are reduced fuel consumption rates and less greenhouse gas (GHG) emission. These environ­ mental gains could be analyzed for their impacts in monetary values. GHG vehicle emissions savings are a direct function of fuel consumption reduction. Table 14 shows the average motor gasoline sale in thousand gallons per day nationally and for the states of Minnesota and Virginia. To estimate the environmental savings due to the introduction of ADS into the traffic network, the following equation framework is proposed: Eq. 5 _ _ _ Environmental benefit FC FC F Em FC F CostTotal Cost Rd GHGGHG type Total Rd ADS GHG typeADS ∑= +p p p p p where: FCTotal = Fuel consumption (gallons) FCCost = Fuel cost (USD per gallon) EmGHG = Emission by GHG type per fuel consumption (lbs per gallon) CostGHG_type = Social cost of greenhouse gas by GHG type (USD per lbs) FRd_ADS = Fuel reduction factor (decimal) due to deployment of ADS features The framework can be decomposed into two parts where the first is the direct fuel consump­ tion cost due to vehicle operations while the second is the indirect environmental impact cost due to GHG emissions. The types of GHG cost can be expanded based on the relevant GHG cost and emissions data available. Table 15 shows the estimated dollar cost per unit for gasoline and sample GHG emissions.

Overview of the Framework Elements 39   Virginia 9,674.7 9,444.5 9,168.0 8,909.1 Location 2016 2017 2018 2019 United States 371,725.6 375,118.3 374,602.1 367,306.0 Minnesota 6,592.5 6,382.7 6,218.3 6,305.1 Table 14. Thousand gallons per day average petroleum sale (motor gasoline). Fuel/Emission Per Units Cost Gasoline Gallon―2019 dollar $2.258 CO2 Metric ton―2007 dollar $12 NOx Metric ton―2007 dollar $4,700 1https://www.eia.gov/dnav/pet/hist/LeafHandler.ashx?n=pet&s=emm_epm0_pte_nus_dpg&f=a. 2https://www.eia.gov/electricity/annual/html/epa_a_02.html. 3https://www.epa.gov/energy/greenhouse-gases-equivalencies-calculator-calculations-and-references. Table 15. Gasoline and GHG environmental costs.1,2,3 Table 16 shows the estimated GHG emissions per unit of fuel consumption by internal combustion engines. The FRd_ADS is the expected percentage of reduction in the fuel consumption due to ADS deploy­ ment. This factor would be very close to the percentage of the ADS­equipped vehicles within the fleet. ADS­equipped vehicles would be fully electric vehicles with no internal combus­ tion engine. Therefore, they would have zero fuel consumption and zero tailpipe emission. IOOs should be aware that ADS­equipped vehicles still add environmental costs relevant to GHG emissions that accompany battery production. However, these costs are not relevant to the transportation network and are expected to be outweighed by the zero emissions of the ADS­equipped vehicles through their lifetime. Step 6—Communicate Outcomes The objective of the framework is to assist transportation agencies in quantifying the safety performance of ADS under various scenarios. The safety performance can be evaluated in con­ cert with other quantitative measures (e.g., costs, operational efficiency, environmental impacts) or qualitative measures (e.g., equity, convenience, competitiveness) to inform the decision­making process. While the framework in this guide can help estimate the safety performance of ADS, the ability to use the results to inform decisions is ultimately the key. As such, there is a need to clearly and concisely communicate the analysis results to various audiences. Emission Per Unit Per Unit Converted NOx 443.80 lb per MG1 443.8 x 10-3 lb per gallon CO2 8.89 kg per gallon 19.6 lb per gallon 1MG: thousand gallons. Table 16. GHG emissions per unit of distilled petroleum.

40 Framework for Assessing Potential Safety Impacts of Automated Driving Systems The framework can serve the needs of both technical and nontechnical audiences. These audi­ ences may include OEMs, technical developers, IOOs, policymakers, community advocates, and regulatory staff. Identifying the proper audiences with which to share results is crucial to meeting policy, funding, and public perception goals. It is important to devise and employ targeted communication approaches and messaging to reach diverse audience groups effectively. Furthermore, within each group, the communications tools and methods must effectively edu­ cate and support the potential wide ranges of expertise and interest. For example, the various divisions within IOOs (e.g., transportation planners, highway designers, traffic engineers, and maintenance and operations groups) may be interested in different aspects of the analysis results. Regardless of the target audience and method of communication, engagement with commu­ nications experts is needed throughout the process. Once the appropriate audience groups are targeted and their specific messages are conceived, engaging stakeholders with continuous feedback loops will help effective communication achieve messaging goals. Communications feedback loops not only inform stakeholders using the International Association for Public Participation model (https://www.iap2.org/page/resources) but also empower them. Communicating with public audiences such as community advocates is crucial to sway public perception on the safety of ADS. Key performance indicators (KPIs) should be selected to demonstrate the progression of education and familiarity with ADS as well as willingness to support it. KPIs can be useful metrics to demonstrate to a less technical audience and can be used to track engagement. Additionally, feedback loops allow real­time strategic and coordi­ nated emergency communication based on input from the entire community. The following sections describe different considerations, methods, measures, and formats for presenting the results of safety analyses to convey the key details to technical and nontechnical audiences. Below is a step­by­step approach to ensuring that communications are well thought out to maximize their effectiveness and likelihood of driving the desired outcomes. Communications Goal and Target Objectives The first step in any strategic communications strategy is to determine the overall goal and target objectives. In this case, the overall goal is to convey the results of the analysis conducted, but there may be multiple subsidiary objectives. For example, one objective might be to persuade certain audiences about potential benefits or refute perceived drawbacks. Transportation agencies likely have multiple target objectives, so it is critical to understand what those objectives are in order to determine the best ways to achieve them. Stakeholder/Audience Analysis To do so, the next step is to conduct a stakeholder analysis to identify the target audiences. For state and local agencies, there will be distinct audiences when communicating about ADS, with different—and potentially competing—needs and interests. These include select “internal” audiences, such as the policymakers and regulatory bodies elsewhere within the affected juris­ dictions, as well as numerous external audiences including OEMs, IOOs, technology developers, and community members (residents and the business community). Some of the more technical audiences, such as OEMs, technical developers, and IOO staff (planners, engineers, and analysts) involved with design, development, and deployment of ADS as well as those involved in transportation project planning, design, development, imple­ mentation, operations, and maintenance, may be interested in the following: • Details of the analysis, including data sources, methods, and assumptions. • Details of the results such as the change in the estimated number of crashes by crash type and severity for the various alternatives.

Overview of the Framework Elements 41   • Economic measures such as the benefit­cost ratio or cost effectiveness to support investment decisions. While the technical audiences may have some degree of familiarity with ADS technology and/or quantitative safety measures (e.g., expected changes in crash frequency and severity), there are significant differences in the level of expertise and experience among these audi­ ences. As such, it is important to clearly define the ADS technology of interest, including the capabilities and limitations. It may even be necessary to define the terms and components asso­ ciated with the ADS. The results should also be prefaced with the specific goals, hypotheses, target crashes, and assumptions of the analysis. The assumptions should clearly indicate the ADS penetration rate and the specific ODD. All of these factors are critical to understanding and correctly interpreting the results. Other audiences, such as policymakers, elected officials, community members and advo­ cates from the residential and business communities, and the media, are equally if not more important in investment and policy decisions as technical audiences. Communicating with public audiences is crucial to swaying public perception on the safety of ADS. These audiences may be interested in the following: • Lives saved, injuries prevented, crashes reduced. • Return on investment. Further, it is unlikely for this audience to be interested in the technical details of the ADS technologies and analysis methods. As such, there is a need to focus more on the safety performance (e.g., expected crashes reduced, injuries prevented, and lives saved). This audience will also be interested in the costs, including the cost to purchase a vehicle with the related ADS technology and the cost to improve or maintain the infrastructure to support the ADS within the ODD. It is also useful to show other scenarios such as how the costs and benefits would change if the ODD is expanded. The stakeholder analysis is also useful in identifying potential champions who can help amplify the results and/or deliver them to audiences who might be more receptive to receiving them from an independent entity instead of from the transportation agency directly. For example, one target audience might be local elected officials, and they may be more likely to pay attention to the analysis results if they receive them from community advocates or hear about them from the media than they would if they received them directly from the transportation agency. Once the stakeholder analysis is complete, it can be useful to revisit the goal and key objec­ tives identified in the first step. Key Themes and Messages, and Stakeholder-Message Mapping With the stakeholder analysis in hand, the next two steps occur simultaneously: develop a suite of key themes and messages, and map those messages to the stakeholder analysis. Dif­ ferent themes or messages may be more relevant, topical, or persuasive to different audiences, so transportation agencies should take care to map out the best themes and messages for each stakeholder group. Communications Plan The final step is to build and implement the communications plan itself, which lays out the details for each of the different types of communications that will be conveyed. Just as different audiences may be more receptive to different messages, they will also likely have needs in terms of from whom they receive communications, when they receive communications, and even how they receive communications. For example, do they prefer to receive information directly

42 Framework for Assessing Potential Safety Impacts of Automated Driving Systems from transportation agencies? Or from their constituents? Or from elected officials? Do they prefer information delivered in a formal or academic style? Or via a presentation? Or more informally via a one­on­one conversation? Is it important for certain audiences to receive the information before or after other audiences? Think critically about how the sequence and different potential rollout schedules might affect your target stakeholders. Continuous feedback loops are critical. Make sure you are constantly in touch with your stakeholders to ensure they understand the message and that you know whether to adjust or adapt your messaging at any point to respond to (or better yet, to anticipate) any confusion, misgivings, or countervailing information.