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.

43   Framework Proof of Concept with MnDOT— Rochester Automated Shuttle Pilot Minnesota Department of Transportation (MnDOT) has many planned automated driving system (ADS) activities. Goals for MnDOT’s ADS activities include understanding the infrastruc- ture needs for scaled autonomous surface transportation and the effects of cold weather condi- tions on ADS technologies. MnDOT wants to understand the limitations of operating under cold conditions in an area with snow and ice during several months of the year and to encourage industry to understand these challenges as well. For example, MnDOT is halfway through a project on a 50-mile corridor of Highway 52 with the goal of mapping technologies to transportation challenges. The corridor runs through a highly heterogenous area, including urban, rural, and suburban sections of highway. The project team discussed state and local safety perspectives and has focused on three safety categories: (1) safety needs, (2) operational services (travel times/reliability services), and (3) multimodal (long-range transit trips). MnDOT also has plans for multiple ADS pilots involving mass transit operations. Early-stage goals for ADS projects include introducing ADS technologies to the public and beginning a dis- course of public engagement and education on emerging technologies in transportation. MnDOT has a particular emphasis on community engagement because investment and funding for the infrastructure owners and operators (IOOs) depend on community support. MnDOT foresees early adoption of ADS technologies at the mass transit level due to established operational design domains (ODDs) and relative ease for public engagement when compared with private passenger vehicles. Furthermore, MnDOT’s strategic plan and keen interest to use key performance indicators (KPIs) complement the intention of the framework. One of the envisioned pilots is the Rochester Automated Shuttle Pilot. The pilot will consist of a low-speed, highly automated shuttle bus in downtown Rochester operated by First Transit. The goal of the project is to operate in an urban area for a year to gather lessons learned on automated vehicle (AV) operations in all weather conditions, educate the public on ADS tech- nologies, and provide mobility solutions through an enhanced ADS transit service to the city of Rochester. First Transit will operate the 12-passenger EasyMile EZ10 shuttle. The Level 4 shuttles will have no steering wheel or pedals and will travel between 12 and 15 mph for the duration of the route. A remote attendant will be on standby ready to take control as necessary. As shown in Figure 12, the proposed pilot route is a six-block by three-block circulator route that operates clockwise on 6th St. SE, 3rd Ave. SW, W Center St., and S Broadway Ave. It will connect the Mayo Clinic Hospital Methodist Campus with hotels, shops, restaurants, grocery stores, and parking lots with proposed stops, as shown in Figure 12. C H A P T E R 4 Direct Application of the Framework

44 Framework for Assessing Potential Safety Impacts of Automated Driving Systems Figure 12. Proposed route for the AV shuttle (Source: MnDOT).

Direct Application of the Framework 45   The scope of the Rochester Autonomous Shuttle Pilot is to assess the expected impacts of the project on the future of transportation safety (which is consistent with the title and scope of NCHRP Project 17-91). The expected safety impacts (benefits or disbenefits) are defined by comparing the expected safety performance with the autonomous shuttle to the expected safety performance without the autonomous shuttle. The extent of the network for the analysis includes the street segments and intersections along the fixed transit route that may be impacted by the autonomous shuttle (e.g., mode shift from walking, biking, or driving to shuttle ridership). It is important to note that the intent of the Rochester Autonomous Shuttle Pilot is not to provide additional public transit or induce some type of mode shift; it is simply to test the AV shuttle in inclement weather and introduce the community to ADS technology. The following is a discus- sion of each step of the framework. Step 1—Identify ADS Application(s) of Interest The selected ADS application to include in the proof of concept is the MnDOT Autonomous Shuttle Pilot. The agency was mainly interested in introducing new transportation technologies to the public to gauge public perception. It was seen more as sowing seed for people to become comfortable and help set the foundation for future deployments. Other goals of this project are to understand the following: • Are there safety benefits and what data can be used to analyze ADS deployments? • How should KPI be defined to monitor ADS technology deployments and what is important to capture? • Infrastructure needs—what needs to change, or what can be left alone? • How can this be a benefit for mobility? • What is the impact of winter weather on ADS applications, specifically the limitations of the technology operating under cold weather conditions, and how can industry be encouraged to understand these challenges as well? The agency wants to understand and consider public expectations while evaluating ADS or other transportation technology deployments. Completing surveys to understand community perception and interest in engagement can better support investment for ADS. It is easier to engage the public in transit applications and for an autonomous transit system if the ODD is well known to most users. As a result, the agency chose to pilot a low-speed autonomous transit shuttle within the central business district of Rochester. This will help the agency understand how the public perceives the application, evaluate technical feasibility, and understand challenges and lessons learned for future deployments. Step 2—Understand the ADS Application The project team worked with MnDOT and the original equipment manufacturers (OEMs) to understand better the specific technology. To understand the functionality of the ADS, the project team identified the specific technology components and sensor suite. The team reviewed the OEMs’ user manuals and online documentation to define the functionality of the technology and identify technology and infrastructure dependencies. To understand the expected market, the team made assumptions on transit ridership (based on discussions with MnDOT) and plans for the autonomous shuttle (e.g., route, number and capacity of shuttles, number and location of stops, transit schedule, and rider fare). This was used to estimate a hypothetical market for the service, including the number of potential riders and hypothetical mode shift from walking, biking, or driving in the surrounding area. To understand potential changes to the expected market over time, the team worked with MnDOT to identify any plans for expanding the service in the future or for increasing ridership/mode shift over time. This will be crucial to assessing the

46 Framework for Assessing Potential Safety Impacts of Automated Driving Systems scale and timeframe of safety impacts in the area. It is important to note that MnDOT and its partners in the ADS shuttle pilot are not anticipating a mode shift, and the intent of the pilot is not to enhance shuttle service. However, the NCHRP Project 17-91 team developed this hypothetical mode shift scenario to demonstrate how the framework could be used to assess related impacts. For any assumptions, the team has documented potential upper and lower bounds for use in scenario planning or sensitivity analysis. To further understanding of the expected market and potential changes over time, the team looked to other similar deployment examples throughout the United States, including those shown in Table 17. Step 3—Define Deployment Scenarios In this step, the project team worked with MnDOT to estimate the penetration rate, define the ODD, and identify limitations of the ADS. The penetration rate is speculative and could follow diverse scenarios since it hinges on several interrelated factors. As such, the team assumed different rates for use in scenario planning or sensitivity analysis. The assumed rates coincide with different scenarios for factors such as reliability of technology, regulatory challenges, consumer acceptance, and willingness to pay. For the ODD, the team worked with MnDOT to define the spatial and temporal extent of crashes that could be impacted by the ADS. The spatial extent includes the fixed route along which the autonomous shuttle will operate. A more extensive analysis could include the surrounding network from which the autonomous shuttle could attract ridership and result in mode shifts. The team requested data and results from existing travel demand, origin-destination, and other planning-level models relevant to this analysis. The temporal extent considered ODD factors such as speed range, weather, and time of day. Some risks and opportunities are described below. A subset of these was explored based on available data. Below are some potential risks. • Challenges for first responders include disabling, accessing, or moving low-speed shuttles; directing traffic; and signaling right of way. • At low market penetration rates, low-speed shuttles could emerge and contribute to new crash types. For example, traditional nonautomated vehicles following shuttles could experience increased risk for rear-end crashes because of the shuttle’s slower speed. Shuttles could result in Operators Location Service Area Local Motors— Olli, IBM National Harbor, MD City streets EasyMile EZ10 Arlington, TX Private compound Navya Ann Arbor, MI Campus streets EasyMile EZ10 MnDOT Private compound EasyMile/CCTA San Ramon, CA City streets EasyMile/Transdev Gainesville, FL City streets Optimus Ride Boston, MA; South Weymouth, MA City streets May Mobility and Quicken Loans Detroit, MI City streets EasyMile/Transdev Babcock Ranch, FL Private compound Table 17. ADS deployment examples in the United States.

Direct Application of the Framework 47   more aggressive and frequent lane-change maneuvers by the following nonautomated vehicles. This could increase the crash risk for the aggregate traffic stream. • Shuttles will operate in dense areas, with a high likelihood of significant interactions with pedestrians, bikes, and other motor vehicles. • Access to vehicle and safety data is limited. The following are potential opportunities. • Positive disruption to urban areas may result from increased mobility and reduced congestion. • Common method to introduce AV technologies to the public may help open the door for more ADS technologies. • Slow speed mitigates many safety concerns and allows for less sophisticated and costly sensors because stopping distances are shorter. • Controlled environments, low speed, fewer regulatory constraints, and fixed routes allow easy testing and deployment. • Crashes with pedestrians may be reduced (e.g., sensors on shuttles can perceive at-fault pedes- trians better than drivers, particularly in unexpected scenarios; however, there is not enough data to statistically prove that these sensors are better than human drivers in most scenarios). • Low-cost public transportation option may be due to a reduction in labor costs and a reduc- tion in capital and operational costs associated with smaller, lower-capacity vehicles. Step 4—Define Safety Goals and Hypothesis The project team worked with MnDOT to document the overall goals of the proof of concept and define specific safety-related hypotheses. Based on initial discussions, one goal of the autono- mous shuttle deployment is to introduce ADS technologies to the public. While the overarching goal is not related to safety, there is an opportunity to explore several hypotheses related to safety. For example, one hypothesis may be that the autonomous shuttle will improve safety in the area by reducing crash frequency and severity compared to existing conditions or compared to a similar scenario using traditional transit bus. The following are more detailed questions related to the hypothesis regarding crash types, crash severity levels, infrastructure, and data. • How will the frequency of certain crash types change in relation to safety? It is anticipated that crashes involving transit vehicles will decrease with the deployment of low-speed shuttles. There is a potential to reduce other vehicle-, pedestrian-, and bicycle-related crashes if these modes shift to the shuttle, which would remove them from the segments and intersections along the route. • How will the severity of certain crash types change? It is anticipated that crash severity would decrease with the use of low-speed shuttles. These shuttles drive at lower speeds and are auto- nomous, which could reduce crash severity levels. • How will the frequency of other crash types change (e.g., those not involving shuttles)? With the deployment of low-speed shuttles, there is potential for other crash types to change and possibly increase. Low-speed shuttles might contribute to aggressive driving and evasive moves by other drivers, thus contributing to crashes. The deployment could also draw more pedestrians from surrounding areas, which could increase exposure at certain intersections. • How will the autonomous shuttles respond to dynamic conditions (e.g., weather, work zones, and roadway lighting)? • How will safety change if the ODD is expanded and/or infrastructure improvements are made? If the shuttles can operate in additional conditions, then there is the potential to expand the safety benefits. • How will crash contributing factors change? There is the potential to change factors related to road-user condition (e.g., distracted, impaired) and behavior (e.g., speeding) if these users shift to using the shuttles.

48 Framework for Assessing Potential Safety Impacts of Automated Driving Systems In defining the hypotheses and related questions, the team documented the expected deploy- ment timeline, which could include multiple timeframes depending on the certainty in deploy- ment and penetration rates. Finally, the team demonstrated how to map these hypotheses and findings to plans, policies, and procedures. For example, the team attempted to answer questions such as, How do the expected safety-related benefits (or disbenefits) map to the state’s Strategic Highway Safety Plan (SHSP), safety goals, and emphasis areas? Step 5—Choose Analysis Methodology In Step 5, the project team obtained data from appropriate data sources, defined metrics and evaluation criteria, and selected an appropriate evaluation method to test the hypotheses devel- oped in the previous step. For the Autonomous Shuttle Pilot, the team followed the data collection plan and worked with MnDOT and other stakeholders to obtain crash, roadway, traffic, and other relevant data for the study area. Data Sources The study area includes the route(s) where the autonomous shuttle will operate as well as some areas of influence adjacent to the route(s). Other data of interest include transit ridership, pedestrian counts/activity, and origin-destination models. The following describes how the team collected these data elements, including the desired level of detail and source(s). Crash Data The desired crash data elements include the location, type, severity, date, time, and contributing factors (e.g., weather conditions, driver condition/behavior, etc.) related to the crash. The sources of information include MnDOT, City of Rochester, and local transit agencies. Roadway Data The desired roadway data elements include the number of lanes, lane and shoulder width/type, median width/type, presence of on-street parking, presence of bike lanes, presence of sidewalks, and posted speed limit. For intersections, the desired data elements include the number of legs, traffic control, presence of turn lanes, and presence of crosswalks and other pedestrian features. The sources of information include MnDOT, City of Rochester, and desktop data collection by the project team. Traffic and Pedestrian Data The desired traffic data elements include annual average daily traffic (AADT) or other measures of traffic exposure that could be used to estimate AADT for the segments within the study area. The desired pedestrian data elements include any pedestrian counts or major pedestrian genera- tors within the study area that could be used to develop estimates of pedestrian exposure at various intersections. This would support certain crash prediction methods. The sources of information include MnDOT, City of Rochester, and Rochester-Olmsted Council of Governments (COG). Transit Data The desired transit data elements include routes, number of vehicles per route, ridership by route, number of stops, and boarding and alighting by stop. The sources of information include local transit agencies. This did not consider drop spots to make sure the pedestrians did not get dropped on one side of the road and had to cross a busy intersection.

Direct Application of the Framework 49   Surrounding Land Use The desired elements for surrounding land use include the zoning and types of businesses within and adjacent to the study area. The intent of this information is to identify potential origins and destinations of transit riders, pedestrians, and bicyclists. The sources of this information are online databases or datasets that the Rochester-Olmsted COG has compiled. The team used the traditional datasets (crash, roadway, and traffic) to understand crash con- tributing factors and establish the baseline safety performance for existing and future conditions assuming the current (traditional) vehicle fleet. These data allowed the team to quantify and assess the safety performance of traditional vehicles and identify the conditions under which these crashes are occurring. These datasets also helped to quantify the number of crashes by type and severity that could be impacted by the autonomous shuttle under different deployment scenarios. The project team used the following method for evaluating the safety impacts: 1. Used the Highway Safety Manual (HSM) Part C Predictive Method and associated safety analysis tools [e.g., Interactive Highway Safety Design Model (IHSDM)], to estimate the safety perfor- mance of existing and future conditions under the current vehicle fleet (traditional vehicles). Safety performance measures include the expected crash frequency by type and severity. 2. Estimated the safety performance of future conditions with autonomous shuttles in the vehicle fleet. This includes assumptions related to the penetration rate, mode shift, and ADS func- tionality as determined from previous tasks. For example, if there is a shift to autonomous shuttles from passenger vehicles, walking, or biking in the surrounding area, this will reduce the exposure that, in turn, will reduce the predicted crashes from the HSM Part C Predictive Method. The Part C Predictive Method does not, however, account for the potential mix of ADS applications in the vehicle fleet. As such, this step also involves assumptions about the potential impacts of autonomous shuttles on specific crash types. For example, estimating the percentage of crashes related to traditional transit vehicles per vehicle-mile can inform the predictions from the Part C Predictive Method. The team documented assumptions and explored the effects of different ranges of assumptions for use in scenario planning or sensitivity analysis. 3. Used the results from Steps 1 and 2 to estimate the expected impacts of autonomous shuttles based on underlying assumptions. MnDOT provided crash data from 2016 through 2020 for the study area, including crashes along the shuttle loop and crashes along roads on the interior of the loop. Some of the variables in the crash data included severity, first harmful event, road condition, weather, and an indicator for intersection-related crashes. Table 18 displays the crash history along the shuttle loop by year and severity. The crash history includes crashes on segments and at intersections. Year Serious Injury Minor Injury Possible Injury Property Damage Only Unknown Severity Total 2016 0 6 10 57 3 76 2017 1 4 3 40 0 48 2018 1 5 5 45 0 56 2019 0 4 12 59 0 75 2020 1 3 2 27 0 33 Total 3 22 32 228 3 288 Table 18. Crash history along shuttle loop by year and severity.

50 Framework for Assessing Potential Safety Impacts of Automated Driving Systems The project team obtained roadway data and surrounding area characteristics for the IHSDM analysis through a desktop data collection effort using Google Earth. Roadway information included alignment type, lane width, median width, median type, number of driveways, pres- ence of on-street parking, and lighting. The number of schools, alcohol sales establishments, and bus stops within 1,000 feet of an intersection were also estimated using Google Earth. In addition to the crash and roadway data, the project team obtained traffic data for the roadways along the shuttle loop and for the roadways that intersect the loop. MnDOT’s Traffic Mapping Application (MnDOT, 2021) provided traffic volume data (AADT) for the majority of roadways. However, there are a few roads that intersect the shuttle loop that do not have AADT values. For those roads, the project team estimated AADT based on the features of the road and compared them against AADT values for similar roads and surrounding roads. Figure 13 displays the AADT values used in the analysis for roads in the study area. The shuttle route is displayed as a dashed red line in Figure 13. Figure 13. AADT values for each road in study area (Source: © 2021 Google modified by the authors).

Direct Application of the Framework 51   Evaluation Method The project team used the IHSDM Crash Prediction Module (CPM) to predict crashes along the shuttle loop for existing conditions, calculate the expected crashes using historical crash data, and predict crashes for two scenarios that involve a shuttle. The two scenarios are described below. Scenario 1 includes adjusting pedestrian activity at the signalized intersections adjacent to the three shuttle stops. Pedestrian volumes are hypothetically expected to increase at the two signalized intersections directly adjacent to shuttle stops and decrease at all other intersections along the shuttle route. As noted, the shift in pedestrian activity is only hypothetical and does not reflect actual changes due to implementation of the shuttle. The project team assumed that technology does not limit functionality in adverse weather conditions, but this may not be a realistic assumption based on potential limitations of the current technology. Scenario 2 includes adjusting AADT based on a potential mode shift from people using personal vehicles to using a shuttle. The project team assumed an AADT reduction of 7% on all roads along the shuttle route and along roads that intersect the shuttle route. The 7% reduction was based on the number of people that two of the current-style shuttles can accommodate when operating 12 hours per day. This reduction in AADT is only a hypothetical future scenario, assuming the shuttle pilot is successful and that there is a demand for more ADS shuttles to expand service. Again, the project team assumed that technology does not limit functionality in adverse weather conditions, which may not be a valid assumption. First, the project team entered crash, roadway, and traffic data for the existing conditions into the IHSDM for each segment and intersection along the shuttle loop. Figure 14 displays the segments and intersections that form the shuttle loop as they appear in IHSDM. The intent of this image is to display how a network is viewed in IHSDM. To run the analysis, the project team made assumptions about characteristics in the study area. First, driveways were classified as “minor commercial,” and on-street parking was classified as “parallel, commercial/industrial/institutional.” Additionally, all left-turn movements at signalized intersections were considered to be permissive. The project team estimated pedestrian crossing volumes at intersections based on the surrounding area characteristics and estimates of pedes- trian volumes based on general level of pedestrian activity from the HSM. Figure 15 displays the estimates of existing pedestrian volumes at intersections in the study area. For the first scenario, the project team adjusted the pedestrian activity based on the location of the shuttle stops and a hypothetical mode shift. Pedestrian activity at intersections directly next to a shuttle stop is expected to increase, while pedestrian activity at all other intersections is expected to decrease. Figure 16 displays the hypothetical pedestrian activity due to the pres- ence of the shuttle. Results Using the data and assumptions, the project team entered the information into IHSDM to predict crash frequency along the shuttle loop for existing conditions and the hypothetical future scenarios. Table 19 displays the predicted crash frequency from IHSDM for the segments for the existing conditions. As shown in Table 19, there are a total of 5.1 predicted crashes per year, 1.5 predicted fatal plus injury crashes per year, and 3.6 predicted property damage only crashes on the four segments for the existing conditions. Broadway Ave. experiences the most predicted crashes compared to Center St., 6th St., and 3rd Ave. Table 20 displays the predicted crash frequency from IHSDM for the intersections in the study area for the existing conditions. As shown in Table 20, there are a total of 43.4 predicted

52 Framework for Assessing Potential Safety Impacts of Automated Driving Systems Figure 14. Road network of the shuttle loop as it appears in IHSDM (Source: IHSDM project output).

Direct Application of the Framework 53   Figure 15. Existing pedestrian activity at intersections in the study area (Source: © 2021 Google modified by the authors). crashes per year, 16.3 predicted fatal plus injury crashes per year, and 27.2 predicted property damage only crashes at the 16 intersections in the study area for the existing conditions. The 2nd St. and Broadway Ave. intersection experiences the most predicted crashes (6.9 predicted crashes per year) compared to the other intersections, followed by the Broadway Ave. and 4th St. intersection (6.5 predicted crashes per year). Using the predicted crashes and crash history for the existing conditions, the project team used IHSDM to calculate the expected crashes for the shuttle loop for the existing conditions. These results can be used to establish a baseline for comparison with proposed or hypotheti- cal future scenarios and to identify locations where ADS technologies could have the largest impact. Table 21 displays the expected crash frequency from IHSDM by segment for the existing condi- tions. The results in Table 21 indicate there are a total of 7.1 expected crashes per year, 1.8 expected fatal plus injury crashes per year, and 5.4 expected property damage only crashes on the four segments for the existing conditions. Broadway Ave. experiences the most predicted crashes compared to Center St., 6th St., and 3rd Ave.

54 Framework for Assessing Potential Safety Impacts of Automated Driving Systems Figure 16. Hypothetical pedestrian activity due to shuttle at intersections in the study area (Source: © 2021 Google modified by the authors). Location Predicted Total Crash Frequency (crashes/yr) Predicted Fatal + Injury Crash Frequency (crashes/yr) Predicted Property Damage Only Crash Frequency (crashes/yr) Segments 5.1 1.5 3.6 Center St. 0.7 0.2 0.5 Broadway Ave. 2.8 0.9 1.9 6th St. 0.7 0.2 0.5 3rd Ave. 0.9 0.2 0.7 Table 19. Predicted crashes for segments along the shuttle route from IHSDM for the existing conditions.

Direct Application of the Framework 55   Location Predicted Total Crash Frequency (crashes/yr) Predicted Fatal+Injury Crash Frequency (crashes/yr) Predicted Property Damage Only Crash Frequency (crashes/yr) Intersections 43.4 16.3 27.2 3rd Ave. and Center St. 2.1 0.8 1.3 Center St. and 2nd Ave. 0.7 0.4 0.3 Center St. and 1st Ave. 1.7 0.7 1.0 Center St. and Broadway Ave. 5.7 2.2 3.5 1st St. and Broadway Ave. 2.4 1.0 1.4 2nd St. and Broadway Ave. 6.9 2.7 4.2 3rd St. and Broadway Ave. 3.1 1.1 2.0 Broadway Ave. and 4th St. 6.5 2.5 4.1 Broadway Ave. and 6th St. 4.5 1.5 3.0 6th St. and 3rd Ave. 0.5 0.1 0.3 6th St. and 2nd Ave. 2.0 0.8 1.3 6th St. and 1st Ave. 1.7 0.6 1.2 3rd Ave. and 4th St. 1.7 0.6 1.1 3rd Ave. and 3rd St. 0.1 0.0 0.0 3rd Ave. and 2nd St. 3.4 1.2 2.3 1st St. and 3rd Ave. 0.4 0.1 0.2 Table 20. Predicted crashes for intersections along the shuttle route from IHSDM for the existing conditions. Location Expected Total Crash Frequency (crashes/yr) Expected Fatal and Injury Crash Frequency (crashes/yr) Expected Property Damage Only Crash Frequency (crashes/yr) Segments 7.1 1.8 5.4 Center St. 0.8 0.2 0.6 Broadway Ave. 4.9 1.2 3.8 6th St. 0.7 0.2 0.5 3rd Ave. 0.7 0.2 0.5 Table 21. Expected crashes along shuttle route segments from IHSDM (existing conditions).

56 Framework for Assessing Potential Safety Impacts of Automated Driving Systems Table 22 displays the expected crash frequency from IHSDM by intersection for the existing conditions. There are a total of 38.6 expected crashes per year, 11.9 expected fatal plus injury crashes per year, and 26.6 expected property damage only crashes at the 16 intersections in the study area for the existing conditions. The 2nd St. and Broadway Ave. intersection experiences the most predicted crashes (9.5 expected crashes per year) compared to the other analyzed inter- sections, followed by the Broadway Ave. and 4th St. intersection (7.4 expected crashes per year). The expected crashes from IHSDM can also be broken out by crash type for either a segment or an intersection. This information can be used to identify crash types with a high percentage of crashes and identify ADS that could positively impact those crash types. It can also help to identify areas of concern if ADS is expected to exacerbate certain crash types. For instance, if a given ADS technol- ogy is expected to reduce angle crashes and potentially increase rear-end crashes, then one could use this table to understand the potential net impacts. If angle crashes are highly represented and rear- end crashes are not, then this might be an acceptable tradeoff. However, if rear-end crashes are highly represented and angle crashes are not, then this might not provide desirable safety outcomes. Table 23 displays the expected crash type distribution for segments in the study area by severity for the 5-year study period for the existing conditions. Table 23 indicates there are more Location Expected Total Crash Frequency (crashes/yr) Expected Fatal and Injury Crash Frequency (crashes/yr) Expected Property Damage Only Crash Frequency (crashes/yr) Intersections 38.6 11.9 26.6 3rd Ave. and Center St. 0.9 0.6 0.3 Center St. and 2nd Ave. 0.5 0.3 0.3 Center St. and 1st Ave. 1.0 0.5 0.5 Center St. and Broadway Ave. 4.6 1.5 3.2 1st St. and Broadway Ave. 2.0 0.6 1.4 2nd St. and Broadway Ave. 9.5 2.4 7.1 3rd St. and Broadway Ave. 1.6 0.5 1.0 Broadway Ave. and 4th St. 7.4 2.3 5.0 Broadway Ave. and 6th St. 5.2 1.1 4.0 6th St. and 3rd Ave. 0.2 0.1 0.1 6th St. and 2nd Ave. 1.2 0.5 0.8 6th St. and 1st Ave. 1.4 0.4 1.0 3rd Ave. and 4th St. 1.1 0.5 0.6 3rd Ave. and 3rd St. 0.1 0.0 0.0 3rd Ave. and 2nd St. 1.7 0.5 1.2 1st St. and 3rd Ave. 0.2 0.1 0.1 Table 22. Expected crashes for intersections along the shuttle route from IHSDM for the existing conditions.

Direct Application of the Framework 57   multiple-vehicle crashes along segments in the study area compared to single-vehicle collisions. Rear-end collisions are the crash type with the highest expected crash frequency (22.5 expected crashes for the 5-year study period) compared to the other crash types. Table 24 displays the expected crash type distribution for intersections in the study area by severity for the 5-year study period for the existing conditions. As shown in Table 24, there are more multiple-vehicle crashes along segments in the study area compared to single-vehicle collisions. Rear-end collisions are the crash type with the highest expected crash frequency (84.5 expected crashes for the 5-year study period) compared to the other crash types. After predicting crashes for the existing conditions, the project team analyzed the two scenarios using the IHSDM CPM to compare the change in predicted crashes between the existing condi- tions and the two scenarios to determine whether crashes are expected to increase or decrease. The two scenarios, previously described, are: • Scenario 1 includes adjusting pedestrian activity at the signalized intersections adjacent to the three shuttle stops. • Scenario 2 includes adjusting AADT based on a potential mode shift from people using per- sonal vehicles to using a shuttle. Crash Type Fatal and Injury Property Damage Only Total Crashes % Crashes % Crashes % Collision with animal 0.2 0.2 0.4 0.2 0.6 0.2 Collision with bicycle 0.4 0.5 0.0 0.0 0.4 0.2 Collision with fixed object 0.6 0.8 3.1 1.7 3.7 1.4 Collision with other object 0.0 0.0 0.2 0.1 0.2 0.1 Other single-vehicle collision 0.5 0.7 0.6 0.3 1.1 0.4 Collision with pedestrian 1.5 1.9 0.0 0.0 1.5 0.6 Total single-vehicle crashes 3.2 4.2 4.3 2.4 7.5 2.9 Angle collision 0.4 0.5 1.2 0.6 1.6 0.6 Driveway-related collision 0.9 1.2 1.9 1.0 2.8 1.1 Head-on collision 0.2 0.3 0.2 0.1 0.4 0.2 Other multivehicle collision 0.3 0.4 1.8 1.0 2.1 0.8 Rear-end collision 5.5 7.3 17.0 9.3 22.5 8.7 Sideswipe, opposite direction collision 0.2 0.2 0.3 0.2 0.5 0.2 Sideswipe, same direction collision 0.4 0.5 5.3 2.9 5.7 2.2 Total multiple-vehicle crashes 7.9 10.4 27.6 15.1 35.5 13.7 Total segment crashes 11.1 14.6 31.9 17.4 43.0 16.6 Table 23. Expected crash type distribution for segments for the 5-year study period for the existing conditions.

58 Framework for Assessing Potential Safety Impacts of Automated Driving Systems Crash Type Fatal and Injury Property Damage Only Total Crashes % Crashes % Crashes % Collision with animal 0.0 0.0 0.0 0.0 0.0 0.0 Collision with bicycle 2.9 3.8 0.0 0.0 2.9 1.1 Collision with fixed object 2.1 2.8 7.6 4.2 9.8 3.8 Non-collision 0.5 0.6 0.3 0.1 0.8 0.3 Collision with other object 0.2 0.3 0.6 0.3 0.8 0.3 Other single-vehicle collision 0.2 0.2 0.5 0.2 0.6 0.2 Collision with parked vehicle 0.0 0.0 0.0 0.0 0.0 0.0 Collision with pedestrian 15.8 20.8 0.0 0.0 15.8 6.1 Total intersection single-vehicle crashes 21.6 28.6 9.0 4.9 30.6 11.8 Angle collision 15.7 20.7 39.3 21.5 55.0 21.3 Head-on collision 2.0 2.6 3.9 2.1 5.9 2.3 Other multivehicle collision 2.3 3.1 27.4 14.9 29.7 11.5 Rear-end collision 19.0 25.1 65.5 35.8 84.5 32.7 Sideswipe 4.0 5.3 6.0 3.3 10.0 3.9 Total intersection multivehicle crashes 43.1 56.8 142.1 77.6 185.2 71.5 Total intersection crashes 64.7 85.4 151.1 82.6 215.8 83.4 Table 24. Expected crash type distribution for intersections for the 5-year study period for the existing conditions. Table 25 displays a comparison of the predicted total crash frequency for the existing conditions, Scenario 1, and Scenario 2 for the segments along the shuttle route. Table 26 displays the predicted total crash frequencies for the intersections. Comparing the existing conditions and Scenario 1, the results indicate that predicted total crashes do not change for segments and do not change dramati- cally for intersections. However, there are slight changes in predicted crashes between the existing conditions and Scenario 1 at individual intersections (e.g., 3rd Avenue and Center Street inter section). In contrast to Scenario 1, the results for Scenario 2 indicate a 4% decrease in predicted segment crashes and a 7% decrease in predicted intersection crashes compared to the existing conditions. The project team compared the predicted crashes broken out by crash type. Table 27 displays the crash type distributions for segments for the existing conditions, Scenario 1, and Sce- nario 2. The results indicate a 19% decrease in total segment crashes between the existing condi- tions and Scenario 1 and a 33.5% decrease in total segment crashes between the existing conditions and Scenario 2. Table 28 displays the crash type distributions for intersections for the existing conditions, Scenario 1, and Scenario 2. The results indicate a 5% increase in total intersection crashes between the existing conditions and Scenario 1 and a 3% decrease in total intersection crashes between the existing conditions and Scenario 2.

Direct Application of the Framework 59   Location Predicted Total Crash Frequency (crashes/yr) for Existing Conditions Predicted Total Crash Frequency (crashes/yr) for Scenario 1 Predicted Total Crash Frequency (crashes/yr) for Scenario 2 Segments 5.1 5.1 4.9 Center St. 0.7 0.7 0.7 Broadway Ave. 2.8 2.8 2.6 6th St. 0.7 0.7 0.7 3rd Ave. 0.9 0.9 0.9 % Crash reduction ― ― 4% Location Predicted Total Crash Frequency (crashes/yr) for Existing Conditions Predicted Total Crash Frequency (crashes/yr) for Scenario 1 Predicted Total Crash Frequency (crashes/yr) for Scenario 2 Intersections 43.4 43.4 40.2 3rd Ave. and Center St. 2.1 2.0 2.0 Center St. and 2nd Ave. 0.7 0.7 0.6 Center St. and 1st Ave. 1.7 1.8 1.6 Center St. and Broadway Ave. 5.7 5.5 5.2 1st St. and Broadway Ave. 2.4 2.3 2.2 2nd St. and Broadway Ave. 6.9 7.1 6.3 3rd St. and Broadway Ave. 3.1 3.2 2.9 Broadway Ave. and 4th St. 6.5 6.4 6.0 Broadway Ave. and 6th St. 4.5 4.5 4.1 6th St. and 3rd Ave. 0.5 0.5 0.4 6th St. and 2nd Ave. 2.0 2.0 1.9 6th St. and 1st Ave. 1.7 1.8 1.6 3rd Ave. and 4th St. 1.7 1.9 1.7 3rd Ave. and 3rd St. 0.1 0.1 0.1 3rd Ave. and 2nd St. 3.4 3.3 3.3 1st St. and 3rd Ave. 0.4 0.3 0.3 % Crash reduction N/A 0% 7% Table 25. Predicted total crash frequency for segments in the study area for existing conditions, Scenario 1, and Scenario 2. Table 26. Predicted total crash frequency for intersections in the study area for existing conditions, scenario 1, and scenario 2.

60 Framework for Assessing Potential Safety Impacts of Automated Driving Systems Crash Type Total Expected Crashes for Existing Conditions (2021−2026) Total Predicted Crashes for Scenario 1 (2021−2026) Total Predicted Crashes for Scenario 2 (2021−2026) Crashes % Crashes % Crashes % Collision with animal 0.6 1.4 0.7 2.3 0.7 2.4 Collision with bicycle 0.4 0.9 0.4 1.3 0.4 1.4 Collision with fixed object 3.7 8.6 4.0 12.9 3.9 13.6 Collision with other object 0.2 0.5 0.2 0.6 0.2 0.7 Other single-vehicle collision 1.1 2.6 1.2 3.9 1.2 4.2 Collision with pedestrian 1.5 3.5 1.5 4.8 1.3 4.5 Total single-vehicle crashes 7.5 17.4 8.0 25.7 7.6 26.6 Angle collision 1.6 3.7 1.1 3.5 1.0 3.5 Driveway-related collision 2.8 6.5 3.6 11.6 3.3 11.5 Head-on collision 0.4 0.9 0.3 1.0 0.2 0.7 Other multivehicle collision 2.1 4.9 1.2 3.9 1.1 3.8 Rear-end collision 22.5 52.3 13.1 42.1 11.9 41.6 Sideswipe, opposite direction collision 0.5 1.2 0.4 1.3 0.4 1.4 Sideswipe, same direction collision 5.7 13.3 3.4 10.9 3.1 10.8 Total multiple-vehicle crashes 35.5 82.6 23.1 74.3 21.0 73.4 Total arterial segment crashes 43.0 ― 31.1 ― 28.6 ― % Change in crashes ― ― −19% ― −33.5% ― Table 27. Predicted segment crash type distribution for the 5-year study period for the existing conditions, scenario 1, and scenario 2. Step 6—Communicate Outcomes For this step, the team worked with MnDOT and its stakeholders to discuss how the results of the analysis could be used in the decision process and options for presenting the results to dif- ferent audiences. Specifically, the team discussed options for communicating results to technical and nontechnical audiences and how MnDOT might approach a typical project to devise and employ targeted communication and messaging effectively to reach diverse audience groups. As demonstrated from the analysis and results, the framework can be used to estimate a change in crash frequency for various scenarios associated with an AV shuttle. These results can be linked to SHSPs and other safety-related plans or policies. Many SHSPs contain empha- sis areas as well as strategies the state can use to accomplish crash-reduction targets in each emphasis area. By supporting the implementation of ADS technologies, such as ADS-equipped shuttles, states can help achieve the crash-reduction goals laid out in their safety plans.

Direct Application of the Framework 61   Crash Type Total Expected Crashes for Existing Conditions (2021–2026) Total Predicted Crashes for Scenario 1 Total Predicted Crashes for Scenario 2 (2021–2026)(2021–2026) Crashes % Crashes % Crashes % Collision with animal 0.0 0.0 0.1 0.0 0.1 0.0 Collision with bicycle 2.9 1.3 2.9 1.3 2.7 1.3 Collision with fixed object 9.8 4.5 10.1 4.5 9.5 4.5 Non-collision 0.8 0.3 0.8 0.4 0.8 0.4 Collision with other object 0.8 0.4 0.9 0.4 0.8 0.4 Other single-vehicle collision 0.6 0.3 0.8 0.4 0.8 0.4 Collision with parked vehicle 0.0 0.0 0.0 0.0 0.0 0.0 Collision with pedestrian 15.8 7.3 15.7 6.9 15.3 7.4 Total intersection single- vehicle crashes 30.6 14.2 31.3 13.8 29.9 14.3 Angle collision 55.0 25.5 67.2 29.7 62.0 29.7 Head-on collision 5.9 2.7 6.5 2.9 5.9 2.8 Other multivehicle collision 29.7 13.8 26.7 11.8 24.3 11.7 Rear-end collision 84.5 39.2 82.1 36.2 74.7 35.8 Sideswipe 10.0 4.6 12.7 5.6 11.7 5.6 Total intersection multiple- vehicle crashes 185.1 85.8 195.1 86.2 178.7 85.7 Total intersection crashes 215.8 ― 226.4 ― 208.6 ― % Change in crashes ― ― 5% ― –3% ― Table 28. Predicted intersection crash type distribution for the 5-year study period for the existing conditions, scenario 1, and scenario 2. Framework Proof of Concept with Virginia Department of Transportation (VDOT)—ADS-Equipped Trucks Along I-81 (VDOT, 2021) The I-81 corridor serves as a critical north-south backbone of the East Coast’s freight network. It is vital to the efficient movement of goods through the state of Virginia. It connects with five other interstates and traverses 21 cities and towns, 13 counties, and 25 colleges and universities between the West Virginia and Tennessee borders. More than 30% of all trucks and nearly 50% of the state’s value of goods are transported along this corridor (AASHTO, 2021). I-81 has the highest per capita truck volume in Virginia. The high percentage of trucks and steep terrain (Figure 17) are concerning as one truck is equivalent to as many as four passenger vehicles in terms of length (Figure 18). The AADT volume along I-81 ranged from 38,600 to 66,700 vehicles per day in 2016. If these numbers are converted to passenger-car equivalents, the AADT values jump significantly to 59,700 to 90,000 vehicles per day.

62 Framework for Assessing Potential Safety Impacts of Automated Driving Systems Source: I-81 Corridor Improvement Plan, VDOT, December 2018 Figure 17. Elevation along the I-81 corridor (Source: VDOT). Figure 18. Truck along an interstate corridor in Virginia (Source: VHB). As a result, the I-81 corridor is beset by significant safety and reliability issues. There are more than 2,000 vehicle crashes every year, and 26% of those crashes involve heavy trucks. This is the highest percentage for any interstate in Virginia. This results in unpredictable travel delay and affects on-time performance of both heavy commercial vehicles and passenger vehicles. I-81 for the most part has two lanes in each direction; when one lane is blocked, there is a 65% reduction in capacity (VDOT, 2021). Factors that contribute to long crash clearance times include lack of capacity, the rolling terrain, lack of reliable detour routes, and the constrained configuration. Given this background, VDOT wanted to explore a scenario where ADS-equipped trucks and vehicles with forward collision avoidance are deployed along the I-81 corridor as part of the I-81 corridor improvement program. The expected safety impacts (benefits or disbenefits) are defined by comparing the expected safety performance with ADS-equipped trucks and vehicles with forward collision avoidance to the existing crash history along the study corridor. The extent of the study corridor for analysis includes I-81 from milepost 110 to milepost 150, which represents an area north of Roanoke (with higher urban volumes and congestion) to south of Christiansburg (with mountainous terrain, steep grades, and truck climbing lanes). Addition- ally, truck crashes also cost the state in terms of time when roads are closed to clear incidents, delayed freight deliveries, fuel costs, and increased emissions.

Direct Application of the Framework 63   Step 1—Identify ADS Application(s) of Interest The selected applications included in the proof of concept are ADS-equipped trucks and vehicles with forward collision avoidance. The project team worked with VDOT to select a route in Virginia that could potentially benefit from various ADS technologies and, in particular, ADS technologies related to trucks. The team sought to identify a route with high truck volumes and select ADS applications that directly impact crashes involving trucks. VDOT indicated that this route is also heavily congested and has rolling to mountainous terrain. The steep grades influence vehicle speeds, particularly for heavy vehicles, and, when combined with heavy con- gestion, can lead to safety concerns. The results from this analysis can help inform strategic plans such as Virginia’s SHSP and Long-Range Plan, which has a 2045 horizon year. Step 2—Understand the ADS Application Application Description ADS-equipped trucks operate without the need for a driver on a predefined set of roads or geographic area and within a specified ODD. Depending on the level of automation, either a driver or the ADS is the fallback for the dynamic driving task. The ADS automatically collects and processes data from onboard sensors and handles the vehicle-to-vehicle (V2V) communica- tions, if available, to perceive the surroundings (such as relevant signage, roadway markings, and nearby obstacles) and identify the appropriate action to perform the driving task. ADS-equipped trucks in global freight operations are expected to dramatically increase soon. Trucks haul nearly 71% of U.S. freight, with a market size of $740 billion per year (Viscelli, 2018). Automated trucks could double the productivity of long-haul trucking, while reducing energy costs. Experts agree that trucks are great candidates for automation due to high proportions of uninterrupted highway driving. Additionally, demand for ADS-equipped trucking benefits vastly outpaces autonomous passenger cars due to high return on investment (ROI) on vehicles and increased industry efficiency. Expected Market Allied Market Research valued the global self-driving truck market at $1 billion for 2020, and it is expected to reach up to $1.7 billion by 2025 with a compound annual growth rate (CAGR) of 10.4% over the 5 years (Chandani and Baul, 2018). North America is expected to account for the majority of the self-driving truck market share, but the Asia Pacific region will likely exhibit the highest CAGR, 14.7%. The top market driving factors will be rising environmental con- cerns, traffic congestion, road safety, and security. Table 29 lists a few examples of ADS-equipped trucks that are commercially deployed or being tested. Provider Details Michelin Partnered with the startup Einride to develop a Level 4 autonomous driving feature designed to haul up to 16 metric tons at speeds up to 50 mph for over 125 miles. Otto In 2018, Otto performed one of the world’s first shipments by a self-driving truck. The pilot was a demonstration of Otto’s exit-to-exit approach, where the driver does the difficult task of getting the truck to the highway where the ADS takes over. Once the truck was on the highway, the driver was not even in the driver’s seat (Walker, 2019). Table 29. ADS-equipped truck examples.

64 Framework for Assessing Potential Safety Impacts of Automated Driving Systems Operational Design Domain Level Timeline Additional Deployment Context • Freeways (both urban and rural). • Operating only in clear and good weather condition (e.g., no rain, snow, etc.). Short term (high disruption). • Conditional automation (Level 3) where driver is fall back for dynamic driving task. • Cooperative adaptive cruise control. • Long-line haul between cities. • Drivers will be necessary, but vehicle will maintain acceleration, braking, and lane assist. • Operating in mixed traffic. • Operating on well-marked roads and well-maintained signage roads. • Freeways (both urban and rural). • All +4-lane divided highways (urban and rural). • Navigate through interchanges and ramps. • Navigate through signals. Medium term (high disruption). • Operating in mixed traffic. • Level 4 automation ADS application where ADS is responsible for the dynamic driving task fallback and achieving appropriate minimal risk conditions. • Potential for remote piloting. • The driver drives the truck to the freeway and then the driverless feature takes over. • Hand-off trailers between human-driven trucks and ADS- equipped trucks near the exits of the interstate highway system at ADS-equipped truck ports. • Vehicle-to-everything (V2X) communications. • Operating on well-marked and well-maintained signage roads. Table 30. ADS-equipped trucks deployment scenarios of interest. Step 3—Define Deployment Scenarios Operational Design Domain Table 30 summarizes the anticipated ODD elements of ADS-equipped trucks for two different predicted timelines, the short term (next 5 years) and the medium term (next 5 to 10 years). In addition, the table outlines the major deployment specifications envisioned for the ADS applications that are expected to impact its safety assessment. The deployment elements are also provided for the two identified timelines. Based on discussions with VDOT, one of the most appealing capabilities of ADS-equipped trucks is the ability to operate in mixed traffic. In an earlier study, VDOT examined the potential for truck-only lanes on I-81 and concluded that this type of infrastructure would be cost-prohibitive (VDOT, 2007). Table 31 provides a summary of the technology specifications and key infrastructure needs pertinent to the envisioned ADS-equipped trucks. Stage of Technology Development Generation (Gen I) Gen I is the first version of the ADS-equipped truck sensor package and the underlying computational algorithms for processing the data. Typically, this package embraces the needed combination of sensors such as forward-facing cameras, radar, ultrasonic sensors, laser scanners,

Direct Application of the Framework 65   and inertial measurement units (gyroscopes and accelerometers) with a priori digital maps (lane-level detail). This package will not have a good object detection capability in low- visibility conditions, limiting the ODD to certain conditions (e.g., light rain, no snow, good lane markings). Similarly, the underlying perception algorithms for processing data can handle the basic computations for the proper functionality of the ADS-equipped truck feature within the ODD and deployment context. These algorithms are at early development and may have more errors than later, more mature technology, leading to lower safety performance and/or lower percentage of time operating in automated mode (high disengagement rate). Sensors and computation algorithms used at this stage are commercially available and currently operate in certain vehicles. Second Generation (Gen II) Gen II is a more advanced stage of technology than first-generation models or systems. In addition to the sensor types included in Gen II, this generation embraces high-fidelity lidar sen- sors and an onboard unit (OBU). All Gen II sensors are newer, more advanced, and accurate and have a longer perception range than the Gen I sensors. A key feature of the Gen II sensor package is providing layers of redundancy to one another. The perception algorithms frequently cross-check the data from different sensors to ensure that no object is left undetected and to eliminate false positives. This sensor package manifests as a mature sensor fusion technology that is able to combine the sensing capabilities of multiple sensors, resulting in more reliable and robust perception Table 31. ADS-equipped trucks key infrastructure requirements. Expected Timeline Vehicle Type Sensor Package Key Infrastructure Requirements Digital Physical Short term (High disruption). Heavy- duty. Gen I. • V2V communications. • Global positioning system (GPS). • High-definition (HD) maps. • Weather data. • Infrastructure data. • Work zone alerts. • Clear lane markings. • Visible signage. • Highly detectable traffic control device. Medium term (High disruption). Heavy- duty. Gen II. • V2X communications. • GPS. • HD maps. • Weather data. • Infrastructure data. • Work zone alerts. • 5G and dedicated short- range communication. • Lane markings visible. • Visible signage. • Highly detectable traffic control device. • ADS-equipped truck port.

66 Framework for Assessing Potential Safety Impacts of Automated Driving Systems with a broad sensing scope. To this end, the underlying perception algorithms for processing the data are more advanced and are capable of performing complex sensor fusion calculations enabling the operation in an expanded ODD and deployment context. Another envisioned key feature of this generation of sensors is integrating V2V and vehicle-to-infrastructure (V2I) com- munication within the vehicles through the OBU. This would provide opportunities for the vehicle to receive real-time dynamic data for weather, work zones, and traffic. The new advanced sensor suite will allow trucks to operate effectively on more road types, such as four-lane divided highways, in more severe weather, and on roadways with imperfect lane markings and signage. Table 32 provides a qualitative assessment of the technology state of the different sensor package generations. The table highlights the key functional and technical differences between the two generations as well as the operational atmospheric conditions. Infrastructure Needs and Impacts The features will assist the driver navigating a highway. The sensors on board the vehicle need to detect infrastructure elements, such as lane markings, barriers, and signs, to determine proper heading and speed. The infrastructure requirements for this feature are largely driven by chal- lenges in human factors, connectivity, and limits to ADS perception technology. To increase functionality and efficiency of ADS-equipped truck features, ADS-equipped truck ports (ATPs) may need to be constructed near interstates. At ATPs, drivers operating locally can swap trailers to automated tractors optimized for highway driving. Likewise, highway- optimized trucks can swap trailers to human-driven trucks for last mile and urban delivery where driverless operations are more complex (Figure 19). The feature uses more advanced technologies compared to existing platooning technologies, including radar, cameras, and dedicated short-range communication, which may have chal- lenges perceiving certain aspects of infrastructure. Vision is predominantly used to detect lane markings and signage; therefore, it is important that they are as machine readable as possible. Scenario Scenario Technology State Comparison Qualitative Assessment of Technology State Operational and Atmospheric Conditions Key Functional and Technical Differences Gen I • Higher-priced vehicles. • Less sophisticated algorithms, making driverless mode active less often. • Communication with other vehicles. • Weather: clear, wind. • V2V communications. Gen II • Lower-priced vehicles. • More sophisticated algorithms, making driverless mode active more often. • Communication with vehicles and roadway infrastructure. • Weather: clear, wind, rain. • Lidar. • V2X communications. Table 32. Scenario technological specifications of ADS-equipped trucks.

Direct Application of the Framework 67   Cameras are an important part of perceiving the road structure and signage and classifying objects. Cameras do not perform well in precipitation and fog and are dependent on ambient light to detect infrastructure components. Therefore, AV deployment of ADS applications will benefit from efforts to make infrastructure more easily perceived by machine vision in a variety of lighting and weather conditions, such as lane markings that are wider, higher contrast, more retroreflective, and well maintained. Risk Assessment Risks • A major risk is navigating the “machine-to-human handover,” when the technology requests to hand back control to the human. Since it is irresponsible for the technology to simply signal to the human, “Here, you take over,” it is evident there must be a transition time following the handover request for the human driver to regain proper situational awareness. • Relatively low numbers of units are sold by truck manufacturers (Viscelli, 2018). • Lateral wandering of ADS-equipped trucks is much narrower than for human-driven trucks. Pavement fatigue in turn increases the risk of hydroplaning (Zhou et al., 2019). • Labor opposition due to job loss. • Training needs on ADS and ATP use. Opportunities • ADS-equipped truck driving is a promising technology that could bring great benefits to society and road users. In fact, the wide benefits achieved by self-driving trucks (e.g., increased hours of operations and road capacity) are expected to be the main reason for expanding the market of this feature more rapidly than other features for passenger cars. • Unlike with cars, there is already high demand for ADS-equipped trucks. Because of the labor savings of autonomy and because trucks are bought as business decisions thoroughly evaluated by fleets, ROI on ADS-equipped trucks is expected to be high (Viscelli, 2018). In 2013, Moran Stanley estimated that ADS-equipped trucks would provide $168 billion in savings. • Implementing ATPs will provide a host of benefits to both industry and drivers that are at risk from automation. ATPs can be built in strategic locations near interstate exits and truck Figure 19. What an ADS-equipped truck port could look like (Source: Adopted from Viscelli, 2018).

68 Framework for Assessing Potential Safety Impacts of Automated Driving Systems parking lots outside congested urban areas. ATPs not only allow trailer switching, but also provide driver facilities and refueling and charging stations. ATPs could facilitate off-peak deliveries to reduce road congestion and cut down on the coordination between shippers and carriers. Efficiency could be greatly enhanced through a ride-sharing style service that matches drivers and freight through an app with real-time pricing, keeping wages and work opportunity high (Viscelli, 2018). • In the future, many trucks with ADS capabilities will likely be electric. In the United States, the transportation sector is responsible for almost 30% of annual GHG emissions (US EPA, 2019). Battery electric vehicles (BEV), however, have been shown to reduce overall GHG emissions and pollution relative to vehicles with an internal combustion engine and could greatly reduce the need for oil (Delucchi et al., 2014; Lattanzio and Clark, 2020). Transferring emissions from the tailpipe to power-generating plants also further centralizes total emissions in the power production sector where measures such as carbon capture and sequestration and a cleaner fuel mix could contribute to reduced overall emissions. Additionally, BEV engines are inherently more energy efficient than internal combustion engines and can increase energy efficiency further by making use of “energy recovery” technology where breaking and unaccelerated motion act to recharge the battery (Delucchi et al., 2014; Man- zetti and Mariasiu, 2015). Also, this feature eliminates the need for a highly skilled driver in following trucks, which could bring shipping costs down when platooning is deployed at a large scale. • ADS-equipped trucks could positively impact other road users by offering safety benefits when applied at a large scale by reducing or eliminating truck driver errors, at least for the portion of the trip that is operated by the ADS-equipped truck. • ADS-equipped trucks will provide significant safety benefits due to changes in operating hours and reduction of human error. ADS-equipped trucks will likely operate during off-peak hours, reducing traffic congestion and its associated crashes due to fewer inter actions with passenger vehicles. ADS-equipped trucks will help reduce driver fatigue and human errors, which are associated with 94% of serious crashes (NHTSA, 2019). Step 4—Define Safety Goals and Hypothesis For the deployment scenario of ADS-equipped trucks, the goal is to reduce the frequency and severity of truck-involved crashes through the use of ADS-equipped trucks (SAE Levels 3 and 4) and supporting infrastructure. The overall hypothesis is that ADS-equipped trucks will improve safety on I-81 by reducing truck-involved crashes during non-adverse weather conditions. The expected change in the number and percentage of truck-involved crashes will depend on market penetration and the ability of the technology to mitigate certain crash types and events, which are explored in the analysis. The questions to evaluate the overall hypothesis are listed below and relate to crash types, crash severity levels, infrastructure, and data. 1. How will the frequency of truck-related crashes change? It is anticipated that ADS-equipped trucks will impact the frequency of truck-related and truck-involved crashes. 2. How will the severity of truck-related crashes change? ADS-equipped trucks will traverse roads differently than human-driven trucks (e.g., different speeds, ability to stay within lane, etc.). The different driving behavior could alter the severity of truck-involved crashes. 3. How will the frequency of non-truck-related crashes change? While it is anticipated that the frequency of truck-related crashes will decrease, the frequency of non-truck-involved crashes could also change. For example, if truck-involved maneuvers contribute to other vehicle crashes, and ADS-equipped trucks can avoid or reduce these types of maneuvers, then there

Direct Application of the Framework 69   is the potential to reduce crashes in which the truck is not one of the vehicles involved in the crash. Conversely, if ADS-equipped trucks can detect and react to situations faster than human-driven vehicles, this could lead to a potential increase in rear-end crashes, particularly if the large trucks limit forward sight distance for following vehicles. This leads to a follow-up question: How can forward collision avoidance in passenger cars mitigate this potential risk? 4. Will safety of ADS-equipped trucks change if the ODD is extended in which ADS-equipped trucks can operate? For example, the anticipated ODD for ADS-equipped trucks is currently higher classifications of roads (e.g., interstates, freeways, etc.). In summary, through the deployment of ADS-equipped trucks, it is hypothesized that the fre- quency and severity of truck-related crashes will be reduced by 5–10%. The hypothesized reduc- tion can be based on previous research or crash reduction goals of a specific agency. While this provides an overview of potential safety impacts, it is important to perform a crash sequencing exercise to think through the contributing factors and precipitating events that lead to a crash. The following are a few examples related to truck-involved crashes along the interstate. 1. Run-off-road: a) Driver of truck is distracted, falls asleep, or is otherwise inattentive and the vehicle drifts off the road. b) Driver of truck is fully attentive and adverse weather contributes to driver losing control or incorrectly navigating and the vehicle leaves the road. c) Driver of truck is fully attentive and sudden congestion leads to an evasive maneuver whereby the driver attempts to avoid the back of queue and the vehicle leaves the road. 2. Rear-end: a) Driver of truck is distracted, falls asleep, or is otherwise inattentive and the truck rear- ends another vehicle. b) Driver of truck is fully attentive and adverse weather contributes to limited stopping dis- tance whereby the driver is not able to stop or slow and the truck rear-ends another vehicle. c) Driver of truck is fully attentive and sudden congestion leads to unanticipated braking whereby the driver attempts to stop, but the truck rear-ends another vehicle. From the anticipated capabilities of ADS-equipped trucks and the above crash sequencing, the project team identified specific opportunities for ADS-equipped trucks to mitigate crashes. For example, ADS-equipped trucks are not expected to operate in adverse conditions, so there is limited potential to mitigate crashes related to sequence 1b; however, ADS-equipped trucks are expected to provide opportunities to mitigate crashes related to sequence 1a and 1c. Similarly, ADS-equipped trucks are not expected to mitigate crashes related to sequence 2b but are expected to mitigate crashes related to sequence 2a and 2c. A similar exercise could be completed for forward collision avoidance in passenger cars. The analyses in Step 5 explore the specific crashes that could be mitigated by ADS-equipped trucks and forward collision avoidance in passenger cars. Step 5—Choose Analysis Methodology Data Sources VDOT provided historical crash data from 2014 through 2020 for Virginia. Variables in the crash data included severity, collision type, road surface condition, weather, and an indicator for truck-related crashes. The project team filtered the data to include crashes in the study area along I-81 from milepost 110 to milepost 150. Table 33 displays a summary of the crashes that occurred along the study corridor by year and severity, Table 34 displays the crashes by col- lision type, and Table 35 displays the crashes by weather condition when the crash occurred. As shown in the tables, crashes are generally increasing throughout the study period with a dip in 2020. Rear-end crashes are the most prevalent type for total crashes, which is consistent with the

70 Framework for Assessing Potential Safety Impacts of Automated Driving Systems Year Fatal Injury Suspected Serious Injury Suspected Minor Injury Possible Injury Property Damage Only Total 2014 1 20 53 7 282 363 2015 2 17 62 10 327 418 2016 4 22 67 11 370 474 2017 5 21 67 12 363 468 2018 3 25 82 17 468 595 2019 4 16 86 7 447 560 2020 7 15 49 10 319 400 Total 26 136 466 74 2,576 3,278 Table 33. Crashes along I-81 from milepost 110 to 150 by year and severity (2014–2020). Collision Type Total Crashes Total Truck- Involved Crashes Truck- Involved Fatal Crashes Truck- Involved Injury Crashes Truck- Involved PDO Crashes Rear end 1,246 330 7 96 227 Angle 156 90 3 23 64 Head on 4 1 0 1 0 Sideswipe―same direction 512 362 0 67 295 Sideswipe―opposite direction 4 2 0 0 2 Fixed object in road 22 2 0 0 2 Non-collision 53 16 0 5 11 Fixed object―off road 859 142 4 32 106 Deer 363 20 0 3 17 Other animal 28 2 0 0 2 Pedestrian 1 1 0 1 0 Backed into 10 7 0 0 7 Other 20 4 7 0 4 Total 3,278 979 14 228 737 Note: PDO = property damage only. Table 34. Crashes along I-81 from milepost 110 to 150 by collision type (2014–2020).

Direct Application of the Framework 71   Weather Condition Total Crashes Total Truck- Involved Crashes No adverse condition (clear/cloudy) 2,492 763 Fog 19 7 Mist 36 9 Rain 510 136 Snow 162 51 Sleet/hail 57 13 Other 1 0 Severe crosswinds 1 0 Total 3,278 979 Table 35. Crashes along I-81 from milepost 110 to 150 by weather condition (2014–2020). input from VDOT and the recurring congestion issues. Rear-end crashes are the second most prevalent type for truck-involved crashes, second only to sideswipe, same-direction crashes. In total, 979 crashes (30% of total crashes along the study corridor) involved a large truck, and 330 rear-end crashes (26% of rear-end crashes along the study corridor) involved a large truck. The majority of total crashes and crashes involving a large truck occurred during no adverse weather conditions (76% of total crashes and 78% of large-truck-involved crashes), which is fol- lowed by crashes occurring during rain (16% of total crashes and 14% of truck-involved crashes). Evaluation Method The project team used the study corridor crash data to analyze two scenarios related to the number of trucks with ADS capabilities and the number of passenger vehicles with forward col- lision avoidance. The hypothetical scenarios are: • Scenario 1: Various percentages of ADS-equipped trucks (5, 25, and 50%) in the fleet with no passenger vehicles equipped with forward collision avoidance. This scenario is expected to impact crashes that involve large trucks as the at-fault vehicle. • Scenario 2: Various percentages of ADS trucks (5, 25, and 50%) in the fleet with various per- centages of passenger vehicles equipped with forward collision avoidance (5, 25, and 50%). This scenario is expected to impact rear-end crashes that involve passenger cars and large trucks where the passenger car is the trailing vehicle. This scenario is expected to build on the crash reduction in Scenario 1 to include a reduction in rear-end crashes due to passenger vehicles with forward collision avoidance. According to FMCSA (2020), from 2016 to 2018, 78.5% of large trucks in rear-end fatal crashes with passenger vehicles occurred when the passenger vehicle rear-ended a large truck; 57.1% of large trucks in rear-end injury crashes with passenger vehicles occurred when the passenger vehicle rear-ended a large truck; and 45.3% of large trucks in rear-end property damage only crashes with passenger vehicles occurred when the passenger vehicle rear-ended a large truck. The various percentages serve as a sensitivity analysis to explore assumptions related to penetration rates and probabilities that a truck is autonomous and, if it is autonomous, that

72 Framework for Assessing Potential Safety Impacts of Automated Driving Systems the autonomous feature is activated and functions properly. Similarly, this serves as a sensi- tivity analysis to explore assumptions related to penetration rates and probabilities that a pas- senger vehicle is equipped with forward collision avoidance, and if it is equipped, that the feature is activated and functioning properly. Crash reductions were calculated for the scenarios using the equation below. The change in crashes is calculated by subtracting the crashes ADS-equipped vehicles can impact from the total number of crashes for the given years. The following equation shows the change in crashes as a percent change, where a positive percent change indicates a safety benefit, and a negative change indicates an increase in crashes. ( )( ) ( ) = − −           ×Percent crash reduction 1 100Total crashes Crashes impacted by ADS feature Total crashes Eq. 6 Results Scenario 1 The project team filtered the crash data to contain only crashes that involved a large truck and only crashes that occurred during clear or cloudy conditions (i.e., no adverse weather conditions). Those crashes were used to estimate the number of potential crashes reduced or eliminated due to various percentages of ADS-equipped trucks in the fleet and no change to passenger vehicles, as shown in Table 36. Table 36 assumes that all ADS features are 100% effective all of the time for the conditions of interest (i.e., truck-related crashes in non-adverse weather conditions). However, it may be more realistic to assume an effectiveness less than 100% to account for ADS features that may not mitigate certain crashes. Results indicate that the greater the percentage of ADS-equipped trucks along the study corridor, the greater the potential reduction of truck-involved crashes during no adverse weather conditions and the greater the potential reduction in the percentage of total crashes. Five percent of ADS-equipped trucks in the fleet result in an expected 1% total crash reduction; 25% of ADS-equipped trucks in the fleet result in an expected 6% total crash reduction; 50% of ADS-equipped trucks in the fleet result in an expected 12% total crash reduction. These results can be used to identify potential safety benefits of ADS-equipped trucks for various penetration rates and as the expected number of trucks with ADS capabilities increases over time. While total crashes may decrease with the onset of ADS-equipped trucks in the vehicle fleet, specific crash types may increase with the use of ADS features, such as rear-end crashes (Petrovic % ADS- Equipped Large Trucks Fatal Injury Truck- Involved Crashes Suspected Serious Injury Truck- Involved Crashes Suspected Minor Injury Truck- Involved Crashes Possible Injury Truck- Involved Crashes Property Damage Only Truck- Involved Crashes Potential Truck- Involved Crashes Reduced % Total Crashes Reduced 5 1 2 6 1 29 38 1 25 3 9 28 5 146 191 6 50 7 19 56 10 292 382 12 Table 36. Truck-involved crash reduction by severity for various percentages of ADS-equipped trucks during no adverse weather conditions (2014–2020).

Direct Application of the Framework 73   et al., 2020). Table 37 displays a potential increase in truck-involved rear-end crashes due to ADS-equipped trucks in the fleet, assuming a 27% increase in rear-end crashes when ADS-equipped trucks are in the fleet. This increase includes rear-end truck-involved crashes where the pas- senger vehicle rear-ends a large truck. As previously mentioned, 78.5% of fatal rear-end crashes involving a large truck occur when passenger vehicles rear-end a large truck; 57.1% of injury rear-end crashes involving a large truck occur when passenger vehicles rear-end a large truck; and 45.3% of property damage only rear-end crashes involving a large truck occur when passenger vehicles rear-end a large truck. Five percent of ADS-equipped trucks in the fleet result in an expected 0.1% increase in total crashes; 25% of ADS-equipped trucks in the fleet result in an expected 0.3% increase in total crashes; 50% of ADS-equipped trucks in the fleet result in an expected 0.5% increase in total crashes. Scenario 2 The project team also analyzed the crash data to estimate potential crash reductions due to both ADS-equipped trucks in the fleet (Scenario 1) and passenger vehicles with forward collision avoidance. Table 38 shows the estimated number of truck-involved rear-end and run-off-road crashes reduced or eliminated due to various percentages of passenger vehicles with forward col- lision avoidance along the study corridor during no adverse weather conditions. These num- bers include rear-end crashes where a passenger vehicle rear-ends a large truck. Run-off-road % ADS- Equipped Large Trucks Fatal Injury Truck- Involved Rear-End Crashes Suspected Serious Injury Truck- Involved Rear-End Crashes Suspected Minor Injury Truck- Involved Rear-End Crashes Possible Injury Truck- Involved Rear-End Crashes Property Damage Only Truck- Involved Rear-End Crashes Expected Increase in Truck- Involved Rear-End Crashes % Total Crashes Increased 5 0.1 0.1 0.3 0.0 1.1 2 0.1 25 0.4 0.7 1.7 0.2 5.3 8 0.3 50 0.7 1.3 3.5 0.5 10.5 17 0.5 % Passenger Vehicles with Fatal Injury Truck- Involved Suspected Serious Injury Truck- Involved Suspected Minor Injury Truck- Involved Possible Injury Truck- Involved Property Damage Only Truck- Involved Expected Decrease in Truck- Involved Forward Collision Avoidance Rear-End Crashes Rear-End Crashes Rear-End Crashes Rear-End Crashes Rear-End Crashes Rear-End Crashes 5 0.3 0.5 1.3 0.2 3.9 6.2 25 1.4 2.4 6.4 0.9 19.7 30.8 50 2.7 4.9 12.8 1.7 39.5 61.6 Table 37. Rear-end truck-involved crashes increase when passenger vehicle rear-ends a large truck by severity for various percentages of ADS-equipped trucks during no adverse weather conditions (2014–2020). Table 38. Rear-end crash reduction where passenger vehicle rear-ends a large truck by severity for various percentages of passenger vehicles with forward collision avoidance during no adverse weather conditions (2014–2020).

74 Framework for Assessing Potential Safety Impacts of Automated Driving Systems crashes that could be the result of drivers trying to avoid a rear-end crash with a truck should be included, but this level of detail is not readily available in the current data (i.e., no information that a passenger car was following a large truck before the vehicle left the road). If that informa- tion was available, the research team would have included run-off-road crashes from 7 a.m. to 7 p.m. along I-81 between milepost 140 and 150, which represent common congested conditions in the Roanoke area. The estimated reduction in truck-involved rear-end crashes due to passenger vehicles with forward collision avoidance (Table 38) is combined with the estimated crash reduction due to ADS-equipped large trucks in the fleet (Table 36) and the estimated rear-end crash increase (Table 37), as shown in Table 39. Results indicate that as the percentage of ADS-equipped trucks and passenger vehicles with forward collision avoidance in the fleet increases, the greater the estimated total crash reduction along the study corridor. These results can be used to quantify potential safety benefits of ADS-equipped trucks and vehicles with forward collision avoidance and quantify how the safety benefits change as more vehicles on the road have ADS capabilities. Step 6—Communicate Outcomes The analysis supports the goals and hypotheses of the safety impacts of ADS-equipped trucks and vehicles with forward collision avoidance. Through the framework process and analysis of the data, the results indicated that truck-involved crashes and rear-end crashes are expected to decrease in frequency with the deployment of ADS-equipped trucks and vehicles with forward collision avoidance. Additionally, safety is expected to continue to improve if the extent of the ODD expands in which ADS-equipped trucks and vehicles with forward collision avoidance can operate. % ADS- Equipped Large Trucks % Passenger Vehicles with Forward Collision Avoidance Fatal Injury Crashes Suspected Serious Injury Crashes Suspected Minor Injury Crashes Possible Injury Crashes Property Damage Only Crashes Total Crashes % Total Crashes Along Study Corridor Reduced 5 5 0.9 2.2 6.5 1.1 32.0 42.7 1 25 1.9 4.1 11.6 1.8 47.8 67.3 2 50 3.3 6.6 18.1 2.6 67.6 98.1 3 25 5 3.2 9.1 27.3 4.7 144.4 188.7 6 25 4.3 11.0 32.4 5.4 160.2 213.3 7 50 5.6 13.4 38.9 6.2 179.9 244.1 7 50 5 6.0 17.7 53.3 9.2 284.9 371.2 11 25 7.1 19.6 58.5 9.9 300.7 395.8 12 50 8.5 22.0 64.9 10.8 320.4 426.6 13 Table 39. Crash reduction by severity for various percentages of ADS-equipped trucks and passenger vehicles with forward collision avoidance during no adverse weather conditions (2014–2020).

Direct Application of the Framework 75   However, to test the hypothesis and related questions, assumptions were made to estimate the safety impacts of ADS-equipped trucks. Regarding the ODD facility conditions, the technology requires dedicated or separated trucking lanes. However, these lanes are not readily found in existing road networks or explicitly identified in road databases. The analysis assumed that the road network for the facilities of interest had dedicated, separated trucking lanes. Another assump- tion relates to the condition of pavement markings needed for the operation of ADS-equipped trucks. At present, pavement markings need to be in excellent condition. The analysis was per- formed under the assumption that the roadways included in the analysis had pavement markings in excellent condition. The roadway databases used for the ADS-equipped truck analysis do not have information about pavement marking condition, which is typical for these databases.