In Search of Severity Dimensions of Traffic Conflicts for Different Simulated Mixed Fleets Involving Connected and Autonomous Vehicles

Research Article In Search of Severity Dimensions of Traffic Conflicts for Different Simulated Mixed Fleets Involving Connected and Autonomous Vehicles Tasneem Miqdady , Roc´ ıo de Oña , and Juan de Oña TRYSE Research Group, University of Granada, ETSI Caminos, Canales y Puertos, Campus de Fuentenueva, s/n, Granada 18071, Spain Correspondence should be addressed to Roc´ ıo de Oña; [email protected] Received 13 February 2023; Revised 25 April 2023; Accepted 8 May 2023; Published 20 May 2023 Academic Editor: Yanyong Guo Copyright ©2023 Tasneem Miqdady et al. Tis is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Tis study aims to estimate the severity of conficts that may arise from the introduction of connected and automated vehicles (CAVs) by examining the vehicle paths generated by microsimulations of mixed feets of human-driven vehicles and CAVs with diferent levels of automation (L1-L4 vehicles). Te study assesses the severity of conficts using a holistic approach that considers three dimensions: (1) proximity to collision, via the time-to-collision (TTC) indicator; (2) potential consequences of a confict, via single surrogate safety measures such as maximum speed (MaxS) and vehicle speed diference (DeltaS); and (3) a combination of both dimensions to assign severity scores, via TTC and velocity vectors. Te study’s fndings suggest that moderate penetration rates of L3 and L4 vehicles (35–55%) show signifcant diferences in the number of trafc conficts with varying TTC values. Additionally, high penetration rates of L3 and L4 vehicles (above 55%) result in lower values of confict consequences measures such as MaxS and DeltaS. Furthermore, the study shows that confict consequences decrease if the follower is a L3 or L4 vehicle. Te study’s fndings also reveal that there is a considerable reduction in high severity conficts when the penetration rate of CAV levels reaches 50%, and the full operation of L4 vehicles results in a 75.5% reduction in high severity conficts. Terefore, this study provides valuable insight into the potential severe conficts during the transition period from manual vehicle operation to full CAV operation. Overall, the study’s fndings highlight the importance of assessing the severity of potential conficts arising from the introduction of CAVs. By considering the proximity to collision and the potential consequences of conficts, the study provides a comprehensive assessment of the severity of conficts. Tis information can inform the development of policies and strategies to ensure the safe and responsible introduction of CAVs into our transportation systems. 1. Introduction Te forthcoming introduction of connected and automated vehicles (CAVs) on roads has motivated researchers to investigate their various implications, such as trafc delay, congestion, fuel emissions, and trafc safety. Although CAV manufacturers have progressed from CAV research to vehicle prototype production within several automation levels [1], the available (behavioral and crash) data can not sufciently clarify the ambiguity surrounding the crash risks involving CAVs. Accordingly, many studies have used the surrogate safety assessment model (SSAM), developed by the Federal Highway Administration, to analyze the vehicle trajectories gathered from a microsimulation platform to determine trafc safety. Several surrogate safety measures (SSMs) (e.g., time-to-collision (TTC), postencroachment time (PET), and deceleration rate) have been applied to estimate the probability of confict. A trafc confict is an evident instance in which two or more road users or vehicles are near each other in terms of space and time to the extent that the risk of collision exists if their movements do not change [2]. In trafc simulation-based studies, trafc conficts can be Hindawi Journal of Advanced Transportation Volume 2023, Article ID 4116108, 21 pages https://doi.org/10.1155/2023/4116108 determined by modeling trafc fow tracking to extract vehicle pathways over time. Accordingly, SSMs have been extensively employed to identify potential trafc conficts when CAVs share roads. In most previous studies, CAVs typically have a high automation level (i.e., L4) [3–7]. However, other studies have also included several levels of automation [8–10]. In general, they found that increasing the penetration rates of CAV can signifcantly reduce the number of potential conficts. Although the impact of CAVs on trafc safety has been widely studied, to the best of the authors’ knowledge, no study has thoroughly assessed the severity of conficts in trafc streams resulting from the progressive introduction of CAVs. Te novelty of this study is its comprehensive analysis of confict severity involving CAVs under diferent simulated mixed feets (i.e., human-driven vehicles (HDVs) and CAVs of diferent levels). In a review conducted by Zheng et al. [11], they highlighted that it is necessary to establish an adequate trafc confict technique for measuring trafc confict severity, applying a sensitivity analysis to select SSMs threshold and the utilization of a multidimensional defnition of severity. Tus, this study considers these two research directions to devise a reliable technique for assessing the trafc confict among CAVs. Te present approach uses three dimensions for analyzing confict’s severity: (1) the proximity to a collision; (2) potential confict consequences; and (3) a combination of proximity and consequences. Te TTC threshold is considered the margin value for serious conficts [12–14]. So, initially diferent TTC thresholds are tested in this study with the introduction of CAVs of various levels on roads. After identifying the key values for diferent TTC thresholds, the study examines the consequences of confict severity using some SSMs, namely maximum speed (MaxS) and vehicle speed diference (DeltaS). Te study then compares these values in various scenarios and types of vehicle interaction to gain insight into the impact of CAVs on trafc safety. Finally, a time-tocollision (TTC) to velocity change at collision (DeltaV) diagram (i.e., TTC/DeltaV chart) is developed for each automation level to derive a confict severity score. Te remainder of this paper is organized as follows. In Section 2, the analysis of trafc safety and confict severity of CAVs and manually driven vehicles reported in the existing literature is discussed. Te study context modeled by Aimsun [15] and the CAV control algorithms and validation processes used are presented in Section 3. In Section 4, the severity analysis and its results are discussed. Finally, Section 5 summarizes the conclusions and limitations of the study as well as the recommended future directions for CAV trafc safety research. 2. Literature Review Tis section presents the SSMs used to identify trafc confict severity. Afterwards, the extent to which the confict severity in the CAV feld can be predicted is discussed. 2.1. SSMs and Confict Severity. Crash rate and severity are direct indicators of trafc safety performance. However, crashes are rare, and data on aleatory events leading to crashes are not always statistically sufcient for studies. Because of this, SSMs are used to identify trafc conficts and estimate their severity by analyzing recorded videos or a real-time analysis [16, 17], and/or trafc simulation outputs. In fact, by extracting vehicle pathways over time and evaluating their proximity and movements, vehicles close to collisions and with jerky movements are considered to be involved in more severe conficts [18]. Most studies that used SSMs as trafc safety assessment tools implied that a substantial correlation exists between serious conficts and crash severity [19–24]. However, deriving confict severity from SSMs as an indicator has been widely debated, and several trafc confict techniques have been developed over the decades. Previous research has proposed several SSM thresholds to delineate risky/nonrisky conficts. Uniform and nonuniform confict severity zones were also created following various trafc confict techniques and SSM indicators [25]. To predict confict severity, time-based SSMs (e.g., TTC [26], PET [27], time-integrated TTC (TIT), and time exposed TTC (TET) [28]), deceleration-based SSMs (e.g., deceleration rate to avoid crash [29], maximum deceleration rate [30], and rear-end collision risk index [31]), and energybased SSMs (e.g., DeltaV, extended DeltaV [32, 33], and confict index [34]) have been used [35]. Accordingly, trafc confict severity has been defned in terms of three diferent types of SSM. Time-based and deceleration-based SSMs defne severity as the proximity with respect to a crash. Tis is the most prevalent indication for studying trafc accidents and confict severity [33]. However, the early decision-making criteria for severe/nonsevere conficts mainly depended on the assessment of human observers by identifying severe events based on their proximity to a collision [19, 23]. Moreover, a time or space threshold that is commonly employed to align severe conficts has multiple assumptions and validation values. Energy-based SSMs defne severity by another dimension: the consequences of the risk resulting from an interaction (confict). Te idea is that high kinematic forces resulting from vehicle interactions considerably afect road users and probably result in severe injuries and fatalities [2]. Over the years, researchers have indicated their high confdence in this type of indicator for predicting crash severity. Carlson [36] attempted to develop models for estimating the probability of injuries or fatalities in a crash based on variables, such as impact speed and vehicle mass; hence, DeltaV was used to predict injuries and fatalities. Evan [37] subsequently ftted several models using DeltaV to predict injuries and fatalities arising from conficts. Nevertheless, because this indicator was not used for trafc confict analysis until its recent incorporation into SSAM [2], the development of new equations was not distinctly pursued. Consequently, the classical Evan models [37] remained in use. Finally, the third defnition of trafc confict severity is related to the concurrent proportioning of values to proximity and propensity dimensions and generating diferent severity levels. Conficts with potentially high consequences 2Journal of Advanced Transportation Table 1: Summary of previous studies about severity within CAV’s analysis. References Data source CAV considered Context Severity dimension Severity measures Sinha et al. [6] Simulation L4 2-Lane motorway Proximity and consequences TTC, Delta S El-Hansali et al. [43] Simulation L4 6-Lane freeway Consequences MaxS, MaxD, MaxDeltaV Rahman et al. [44] Simulation L1, L2 Arterial (61.15 km) Proximity and consequences TET, TIT, TERCRI, LCC, and NCJ Zhang et al. [45] Simulation L4 4-Lane freeway (7 km) Proximity and consequences TET, TIT, TERCRI, and LCC Laureshyn et al. [32] Video analysis — Urban intersection Proximity, consequences and levels of severity T, DeltaV, extended DeltaV (T/ DeltaV) Souleyrette and Hochstein [38] Simulation — Expressway intersections Levels of severity TTC/MaxDeltaV van der Horst and Kraay et al. [39] Manual confict’ observation — Various Levels of severity TTC and speeds at confict Sinha et al. [46] Field data (crash data) L4 Urban network Consequences Machine learning classifers Chen et al. [47] Field data (crash data) L4 Urban network Consequences Machine learning classifers TTC: time-to-collision, DeltaS: diference in vehicle speeds as observed at tMinTTC, MaxS: maximum speed of either vehicle throughout the confict, MaxD: maximum deceleration of the follower vehicle, DeltaV: velocity change at collision, MaxDeltaV: maximum DeltaV value of either vehicle in the confict, TET: time-exposed-time-to-collision, TIT: time-integrated-time-to-collision, TERCRI: time exposed rear-end crash risk index, LCC: lane changing confict, NCJ: number of critical jerks, T: the expected time for the second (latest) vehicle to arrive at the confict point, and extended DeltaV: models of the integration of T/ DeltaV data. Journal of Advanced Transportation 3 and those that are observed close to the occurrence of crashes are found to have a high probability of severity during the interaction [32]. In the past, a simple human decision-making approach was employed to identify two zones distinguishing severe conficts from the rest of the conficts considering only the proximity threshold value of time. Subsequently, the International Committee on Trafc Confict Techniques contributed to the development of several confict techniques that aided in understanding crash occurrence and its potential severity manually (by observation). Te objective was to establish severity in terms of several levels instead of simply splitting it into two categories (severe/nonsevere) [25, 32, 38]. Ten, the levels were validated by studies conducted abroad. In the Dutch technique (i.e., DOCTOR), the conficts in which speed is high and TTC is less than the threshold value are as deemed severe [39]. In addition, both DOCTOR and the Canadian trafc confict technique [19] incorporate a subjective assessment in which a score (ranging 1–5) determines the probable confict consequences based on evasive action, maneuvering, observed speed, and objective nearness-in-time indicator. Te Swedish trafc confict technique [21, 40–42] considers both the proximity in time and speed at which the confict occurs to indicate severity and refect the potential consequences implicitly. Equidistant parallel severity zones were established by dividing the resulting scores into several levels. Te indicators used for deriving the severity levels were varied (e.g., vehicle speed and distance from a confict site, required deceleration, and friction coefcient) [25]. Moreover, several proximity-to-collision thresholds have been proposed in trafc confict technique research [33]. Other approaches have been employed by other researchers to indicate severity levels. For example, Souleyrette and Hochstein [38] developed an assessment score by defning and adding TTC and DeltaV scores. Ten, some severity lines were ftted by drawing contours for equal assessment score areas. Similarly, Laureshyn et al. [32] incorporated the minimum time leading to an accident and DeltaV in a fgure, thus overftting the extended DeltaV values as severity lines for determining severity levels. 2.2. CAV Crash/Confict Severity. Table 1 provides a summary of previous studies that discussed the severity terms and trafc confict techniques, especially, those considering CAVs in their analysis. Tere are some undergoing CAVs’ tests on public roads in several locations in the United States. In those cases, some studies are able to analyze CAVs’ crash severity based on real data. Sinha et al. [46] conducted a detailed safety analysis using the data from the California Department of Motor Vehicles (2014–2019). Te reported data were used to develop various automated vehicle crash severity models that focused on the injuries for all crash types. However, owing to insufcient data on crashes involving CAVs, the factors that contribute to the severity of a CAV’s crash are not well defned. Nevertheless, various machine learning approaches have been used to better understand CAV crash severity. Chen et al. [47] used a similar approach and found that among all the tested classifers, Xtreme gradient boosting, a decision tree classifcation model, performs better in detecting injuries occurring in CAV crashes. Teir fndings show that if two automated vehicles crash at an intersection or are under adverse weather conditions (e.g., fog and snow), the severity of the crash signifcantly increases. Furthermore, crashes resulting in injuries are more likely to occur in locations with various land use patterns. Diverse land use (e.g., residential, commercial, and public) results in mixed trafc behaviors and changes in regional trafc fow, substantially afecting trafc safety. By contrast, as researchers extensively employ SSMs to understand the safety implications of new trafc designs and alternative safety remedies better, modeling the safety consequences of CAVs and their interactions with HDVs is a relevant application of SSMs. In addition, owing to the limited introduction of CAVs, trafc microsimulation outputs have been used to produce SSMs rather than analyzing videos. Both proximity and consequences dimensions have been used to analyze the severity of CAV conficts. Several proximity SSM indicators have been employed, with TTC being the most prevalent indicator. TIT and TET have been also widely employed in parallel with TTC [44, 45, 48]. By contrast, the distributions of emergency braking [49], rearend collision risk index [44, 50, 51], sideswipe (lane-change conficts) trafc condition [51], and time exposed rear-end crash risk index [51] are all examples of deceleration-based SSMs for evaluating CAV trafc safety [51]. Other surrogate safety indicators, such as standard deviation of speed [51, 52], MaxS, and DeltaS [43, 46, 53], have been used as consequence indicators to assess CAV safety implications. However, to the best knowledge of the authors, no studies have combined all the severity dimensions in CAV trafc safety analysis. In most previous studies, CAVs and HDVs were assessed using the same SSMs and thresholds (e.g., TTC �1.5 s), and no specifc values were considered for CAVs’ confict analysis [5, 43, 45]. By contrast, some researchers suggest that in dealing with CAVs, the default TTC value should be reduced because of their faster reaction times and shorter headways. For instance, Morando et al. [4] tested the ensuing conficts of L4 vehicle penetration using three TTC thresholds: 1.50 s for any confict involving HDVs and two lower values (i.e., 1.00 and 0.75 s) for L4–L4 interactions. Tey indicated that the TTC threshold is an important factor in demonstrating the beneft of CAV introduction in terms of safety. Gu´ eriau and Dusparic [8] and Weijermars et al. [54] proposed 0.75 s for determining conficts involving CAVs. By contrast, Virdi et al. [7] used 0.50 s, and they claimed that regarding their assumption that the headway kept by CAVs is reduced to one-third, then the threshold defning the confict should be also proportionally reduced. Evidently, to distinguish between severe and nonsevere safety critical events (conficts), a sufcient threshold level must be defned. Te defnition of this value is a current challenge concerning conficts involving CAVs that must be scientifcally addressed. 4Journal of Advanced Transportation tAs regards the results of previous studies analyzing trafc confict severity in CAV areas, Rahman et al. [44] used TTC-derived measures (e.g., TET and TIT) as proximity indicators in addition to evasive action indicators (e.g., number of critical jerks and time exposed rear-end crash risk index) as consequence indicators in estimating trafc confict severity when L1 and L2 vehicles enter a trafc stream. Te results reveal that CAV penetration exceeding 60% signifcantly reduces the confict severity at arterial segments and intersections. Sinha et al. [6] studied several SSMs (e.g., TTC, PET, MaxS) by analyzing their distributions at diferent penetration rates and estimating the corresponding crash rates to assess the confict severity for L4 introduction. Teir fndings showed that trafc safety improves, and confict severity and crash rates decrease when the roads are fully operated with L4 vehicles. However, the conficts involving HDVs did not decrease in terms of frequency and severity. ElHansali et al. [43] investigated trafc safety by comparing HDVs and L4 vehicles operating independently at a freeway section (i.e., 100% HDVs vs. 100% L4 vehicles). Contrary to expectations, greater values of severity indicators were observed in the case of L4 vehicles than in the case of HDVs. For example, a higher MaxS was obtained for either vehicle type during conficts, and a higher MaxD was observed when only L4 vehicles occupied the road. Zhang et al. [45] conducted a study that focused on roadway confguration. Using proximity and consequences indicators (i.e., TTC, TIT, TET, time exposed rear-end crash risk index, and lane-changing conficts), they investigated the safety of lanes dedicated for L4 vehicles with diferent penetration rates. Tey emphasized that establishing even one exclusive lane could increase safety because the confict severity was signifcantly reduced in terms of longitudinal and lateral movements. 3. Methodology Aimsun [15] has been used for microsimulation to estimate trajectories for the diferent types of vehicles considered. Subsequently, SSAM [18] was applied to extract surrogate safety indicators for the severity estimation process. 3.1. Study Context. As a test corridor, the study area of a three-lane two-way motorway segment (20.27 km of GR30, an important road leading to Granada City, Spain) (illustrated in Figure 1) was modeled using Aimsun Next. Te geometric design characteristics of the segment (e.g., road profle, curves, and lane detailing) were introduced using an imported Open Street Map of the segment. Te chosen segment has 14 on-ramps and of-ramps and two major entry points. Trafc fow data were gathered using nine detectors installed in the area by the General Trafc Directorate (Direcci´ on General de Tr´ afco (DGT)). Te detectors register instantaneous speeds, trafc volumes, and vehicle type distributions (heavy vehicles vs. passenger cars) at 15-min intervals. Te imported fle data from the DGT sensors for validation were selected for a regular day (Tuesday) and of-peak hour (10:00-11:00 am) because this study modeled a freefow condition. Te average instantaneous speed range was 83–118 km/h, and the trafc count was recorded every 15 min: 547−3570 pc/h and 89–260 hv/h were registered for the GR-30 northbound vehicles, and 809–3281 pc/h and 93–499 hv/h were registered for the GR-30 southbound vehicles. 3.2. Microsimulation Model and Scenarios. Aimsun was selected to calibrate the diferent automation levels of vehicles (from L0 to L4) because it provides specialized tools for CAVs. V2X extension was employed to model the connectivity between vehicles. Te proposed analytical period for the microsimulation is 1 h; however, for trafc validation (volume and speed), this period was broken down to 15-min intervals to refect the real trafc data recorded by the DGT’s detectors better. Trafc operation data are generated using a small time step (i.e., 0.1 s, following previous studies [4, 5] to increase simulation accuracy and reduce the risk of losing vehicle movement details). Te warm-up time was set to 18 min following Wunderlich et al. [55] (based on the road section length and the average speed of vehicles). Furthermore, the model operations were calibrated and validated following the modeling guidelines of Roads and Maritime Services [56]. Miqdady et al. [10] provide more details about this step. After checking the validity of the modeled network, the “car-following and lane-change models” of Gipps [57, 58] that are variants of each travel condition are calibrated to fall within the proposed stochastic dynamic envelopes of CAVs. For instance, CAVs are supposed to have short reaction times, accept short gaps, cooperate in lane changes, etc. Tables 4 and 5 (in Appendix A) show all the parameters that are afected by diferent levels of vehicle automation in the car-following and lane-changing models of Gipps based on previous research [4, 5, 8, 43–45, 54, 59] and logic. Te parameter defnitions are summarized from the Aimsun Granada GR-30 Spain Figure 1: Study area: GR-30 roadway segment in Granada (Spain). Journal of Advanced Transportation 5 user manual. Te values summarized in Tables 4 and 5 are the inputs for the microsimulation models. Tey are provided as means and standard deviations and are normally distributed as suggested by Gipps models for both passenger cars and heavy vehicles. Te analysis attempted to cover a gradually introduction of CAVs with various feet mixes that the real world may encounter. Accordingly, as justifed in Miqdady et al. [10], nine mixed feet scenarios with diferent CAV penetration rates were suggested. Table 2 lists the combinations of HDVs and vehicles with diferent automation levels (L1–L4) in each scenario. 3.3. Severity Analysis. To evaluate the trafc confict severity in the simulated scenarios (i.e., the potential market introduction scenarios of CAV), this study considered the three questions presented by Laureshyn et al. [32]: (i) How can the proximity to a crash be measured? (ii) How can the severity of the consequences of a potential crash be measured? (iii) How can both dimensions be merged? A few studies have analyzed the extent of confict severity in the CAV context [6, 44]. However, they have neither considered all dimensions of severity combined nor analyzed all levels of automation. For the nine proposed scenarios, several SSM indicators were applied to determine both proximity and consequence dimensions at each confict. Te following section illustrates how this study addresses the previous research questions. Te detailed framework is illustrated in Figure 2. Te next section explains the approach followed to explore each severity dimension and presents the results obtained after applying these approaches. 4. CAV Severity Dimensions 4.1. Proximity Treshold. Te most widely used indicator for investigating trafc proximity and confict severity in HDV and CAV confict analysis is TTC [35]. Tis indicator is defned as “the time that remains until a collision could occur if two successive vehicles maintain a speed diference” [28]. It is given by the following: TTCi(t) � xi−1(t) − xi(t) − li−1 vi(t) − vi −1(t),if vi(t)>vi −1(t), ∞,if vi(t)≤vi −1(t), ⎧⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎩ (1) where TTCi(t) denotes the TTC value of the following vehicle, i, at a time instant; t,x, and vdenote the time, position, and velocity of the vehicles, respectively; and l i−1 represents the length of the leading vehicle. A small TTC value indicates a high risk of collision at a given time instant. To assess the severity of vehicle-following events, a TTC threshold must be defned to distinguish between severe and nonsevere conficts [60]. Setting an universal TTC threshold for assessing confict severity has become a matter of contention, particularly in the case of CAV introduction. A review of previous research reveals that several thresholds ranging 0.9–5.0 and 0.5–1.5 s have been proposed for various HDV trafc and driving conditions and for CAV scenarios, respectively [60]. Although this study analyzes trafc safety for CAV introduction scenarios, a unique value (i.e., 1.5 s) is proposed for conficts involving HDVs or vehicles with low automation (L1 and L2) as follower vehicles (low CAVs, LCAV). Te most commonly used value for HDV is 1.5 s [18]; and it is also the default value used in the SSAM. Sensitivity analysis was conducted to defne a reasonable threshold for conficts where a high level of automation vehicle (L3 and L4) is the follower (high CAVs, HCAV). Five diferent values (0.5, 0.75, 1.0, 1.25, and 1.5 s) were examined for the TTC threshold under each scenario to emphasize the appropriate value under various circumstances. Table 3 summarizes the number of conficts when applying the diferent TTC values to determine whether there are signifcant changes by using one-way analysis of variance for each scenario. Te changes resulting from applying any value and the base value (1.5 s) are listed in Table 3. Table 3 shows that (i) TTC does not present a signifcant infuence on the number of conficts at scenarios with low penetration rates of HCAV (scenarios D or below) (ii) TTC presents a very signifcant infuence on the number of conficts at scenarios with high penetration rates of HCAV (scenarios G or over) (iii) At intermediate scenarios (E or F), representing moderate penetration rates of HCAV, the number of trafc conficts starts to present signifcant differences if the TTC value is below 1.0 s. Tese results emphasize the importance of using diferent TTC values to obtain a reliable assessment of trafc safety related to high penetration of HCAV. Moreover, the results verify the theoretical vision of CAV introduction: when CAV penetration rate is high, trafc fow improves by achieving more harmonized speeds and by reducing reaction times that probably have a direct efect on the TTC threshold. Tese results agree with the values suggested in previous studies. Morando et al. [4] used two TTC values (0.75 and 1.0 s) to identify conficts involving CAVs; both values were assumed to be appropriate. Other studies used 0.75 s as the TTC value [8] for fxed conficts with CAV participation, and other studies reduced this threshold to 0.5 s [7, 61]. Papazikou Table 2: Te studied mixed feets’ simulated scenarios. Scenarios HDV (%) L1 (%) L2 (%) L3 (%) L4 (%) A 100 0 0 0 0 B 75 10 10 5 0 C 50 10 25 10 5 D 40 15 20 15 10 E 20 20 25 20 15 F 5 10 30 30 25 G 0 0 10 40 50 H 0 0 0 25 75 I 0 0 0 0 100 6Journal of Advanced Transportation et al. [61] claimed that CAVs operating with assertive driving styles could lead to diferent circumstances resulting in a lower TTC threshold. 4.2. Severity Consequences Indicators. Te proximity to a collision that results in a slight crash must not be equated to a crash with a potentially severe injury. Terefore, the severity measured by the potential consequences of a crash must be accounted by some other means [32]. Several SSMs can be used to extract the dynamic consequences of a confict [18, 35]. Following previous studies [42, 62], this research uses MaxS and DeltaS to measure the resulting severity of conficts related to diferent types of vehicles (HDVs and L1, L2, L3, and L4 vehicles). Te former is defned as the maximum speed of any of the vehicles throughout the confict, whereas the latter is the diference in vehicle speed (i.e., the diference in the velocity of vehicles in confict) observed at the minimum value registered for TTC. Both indicators are outputs of the SSAM and simulate the resulting dynamics. Diferent vehicle interactions can result in varied trafc fow dynamics, and consequently, the severity levels difer. At the end, high MaxS and DeltaS values indicate that the conficts result in high severity. Te variations in MaxS and DeltaS of diferent vehicles involved in a confict within diferent trafc feet scenarios are shown in Figure 3. For simplicity and clarity in presenting the results, L1 and L2 vehicles are grouped as low CAVs and L3 and L4 vehicles as high CAVs, shown as LCAV and HCAV in Figure 3, respectively. Te shown values (of MaxS and DeltaS) are the mean values of 15 runs in each scenario. Te blue-yellow-red scale indicates the increase in severity towards the red color. Lastly, in each fgure, the values are categorized by the follower vehicle in the confict: -HDV, -LCAV, and -HCAV, indicating that the follower vehicle is a HDV, LCAV, and HCAV, respectively. Figure 7 in Appendix B shows an example of the microsimulation results as frequency distributions for MaxS and DeltaS. Tese distributions show also the heat maps’ values (i.e., the mean values exhibited in Figure 3). Examining Severity dimensions among the fleet mixes Microsimulation scenarios (Extracting vehicles trajectories) SSAM analysis (Extracting SSMs) Testing conflict consequences Testing proximity Testing conflict severity score (CAV traffic conflict technique) One-way ANOVA to test conflicts resulted by different TTC threshold (0.50 ,0.75, 1.00, 1.25, & 1.50 s) Heat maps of MaxS and DeltaS by scenario and vehicle interaction Obtaining TTC score: inflection points from the TTC cumulative distribution at pure vehicle type simulation (HDV, L1, L2,L3, or L4 vehicle, exclusively). Establishing MaxDeltaV score. Severity overall scores charts based on the sum of TTC score and MaxDeltaV score. Classification of conflicts by severity score at the simulated scenarios. (i) (ii) (iii) (iv) Figure 2: Framework for simulation-based trafc confict severity estimation for CAV. Journal of Advanced Transportation 7 Regarding the confict consequences extracted at the diferent scenarios, Figure 3 shows that (i) Te higher MaxS during conficts is typically observed in scenarios in which the penetration rate of HCAV is from low to moderate (less than 55%, or scenario F) (see Figure 3(a)) (ii) By contrast, high penetration rates of HCAV (scenarios G, H, and I) result in lower MaxS during conficts (see Figure 3(a)) (iii) Similar conclusions could be obtained from DeltaS’s results in Figure 3(b) Sinha et al. [6] reported a similar pattern. Tey obtained low crash rates and fat distributions for DeltaS values as the penetration rates of L4 vehicles increased. Rahman et al. [44] observed, using other surrogate safety indicators (e.g., TET, TIT, number of critical jerks, and time exposed rear-end crash risk index), that the increase in the penetration rate of vehicles with low automation levels (i.e., L1 and L2 vehicles) decreased the confict severity. Tey found that the highest reduction in severity was achieved when the penetration rate was 100% CAV. By contrast, the reduction was insignifcant when the penetration rate was less than 40%. Table 3: Sensitivity analysis of diferent values of TTC threshold for HCAV (-L3 and -L4 vehicles). Scenario TTC threshold for HCAV No. of conficts % Change A(0)∗— 3251 — B(5) 0.50 2636 −0.69 0.75 2637 −0.68 1.00 2637 −0.60 1.25 2640 −0.56 1.50 2655 — C(15) 0.50 1671 −6.08 0.75 1675 −5.86 1.00 1697 −4.62 1.25 1724 −3.11 1.50 1779 — D(25) 0.50 1131 −11.10 0.75 1137 −10.61 1.00 1156 −8.40 1.25 1200 −5.64 1.50 1272 — E(35) 0.50 890a∗∗ −16.85 0.75 900a −15.91 1.00 935a −12.69 1.25 980a,b −8.50 1.50 1071b — F(55) 0.50 628a −31.22 0.75 648a −28.29 1.00 709a,b −22.36 1.25 770b −15.66 1.50 913c — G(90) 0.50 255a −66.13 0.75 298a −60.46 1.00 415b −44.88 1.25 528c −29.99 1.50 754d — H(100) 0.50 149a −79.03 0.75 198b −72.02 1.00 341c −51.91 1.25 467d −34.06 1.50 709e — I(100) 0.50 133a −82.79 0.75 192b −75.12 1.00 365c −52.67 1.25 517d −32.99 1.50 771e — ∗Te value in ( ) denotes to the percentages of HCAVs in the scenario. ∗∗For each value containing a, b, . . ., letter in a scenario (in the no. of conficts column), it denotes values of statistically signifcant diferences (p<0.05). Two or more values with the same letter denote a homogeneous subgroup. Note. TTC threshold �1.5 s is established when the follower vehicle is a HDV or a LCAV (L1 or L2 vehicle). 8Journal of Advanced Transportation Vehicles involved -HDV MaxS (m/s) 20 22 21 19 23 0 0 0 0 0 2124232127 0 0 0 0 1122212222 0 0 0 HCAV-HCAV LCAV-HCAV HDV-HCAV 0 5 10 15 20 25 GH IEFCBDA Scenario -LCAV MaxS (m/s) 0 0 0 21 20 22 18 12 000 24 16 20 22 23 14 00 0011 23 19 21 20 8.7 Vehicles involved HCAV-HCAV LCAV-HCAV HDV-HCAV 0 5 10 15 20 25 BCDEFGHIA Scenario -HCAV MaxS (m/s) 0 0 0 11 8.6 11 8.9 7.6 0 0 0 6.2 9.6 11 13 15 8.4 0 0 0 9.1 16 19 17 7.1 7.7 0 Vehicles involved HCAV-HCAV LCAV-HCAV HDV-HCAV 0 5 10 15 20 25 BCDEFGHIA Scenario (a) Figure 3: Continued. Journal of Advanced Transportation 9 Table 6: Te assigned TTC score by vehicle type. TTC score HDV L1 L2 L3 L4 Tresholds Sample size (%) Tresholds Sample size (%) Tresholds Sample size (%) Tresholds Sample size (%) Tresholds Sample size (%) 0 4.0 <TTC ≤5.0 30.0 4.2 <TTC ≤5.0 28.9 4.2 <TTC ≤5.0 30.4 4.3 <TTC ≤5.0 29.9 4.3 <TTC ≤5.0 32.8 1 2.5 <TTC ≤4.0 26.9 2.5 <TTC ≤4.2 31.9 2.5 <TTC ≤4.2 31.1 2.6 <TTC ≤4.3 33.6 2.6 <TTC ≤4.3 31.1 2 1.5 <TTC ≤2.5 27.6 1.0 <TTC ≤2.5 32.4 1.0 <TTC ≤2.5 32.3 0.75 <TTC ≤2.6 31.5 0.75 <TTC ≤2.6 31.5 3 TTC ≤1.50 15.3 TTC ≤1.0 6.6 TTC ≤1.0 6.1 TTC ≤0.75 4.8 TTC ≤0.75 4.4 16 Journal of Advanced Transportation 1 2 TTC score 1 23 4 23 4 5 01234 MaxDeltaV score (a) MaxDeltaV sub-score TTC sub-score 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4 4 4 5 5 5 5 55 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 1234 (b) Figure 8: Conceptual illustration of conducting the overall severity score: (a) overall score by regions and (b) step-graded lines from the subscores. HDV 1 2 3 4 5 6 123450 TTC (s) 0 20 40 60 80 100 120 MaxDeltaV (Km/hr) y = 27.273x - 76.364 y = 27.273x - 46.364 y = 25.532x - 7.6596 y = 24.194x + 29.032 y = 23.377x + 63.117 (a) LCAV (L1/L2) 0 20 40 60 80 100 120 MaxDeltaV (Km/hr) 123450 TTC (s) y = 27.778x - 78.889 y = 24.324x - 31.622 y = 25x - 5 y = 25.862x + 20.69 y = 25x + 55 (b) HCAV (L3/L4) 123450 TTC (s) 0 20 40 60 80 100 120 MaxDeltaV (Km/hr) y = 29.126x - 85.631 y = 23.196x - 25.979 y = 24.742x - 3.7113 y = 26.786x + 16.071 y = 24.896x + 55.519 (c) Figure 9: Severity scores (SS) for diferent types of vehicles: (a) for HDV, (b) for L1 & L2 vehicles (LCAV), and (c) for L3 & L4 vehicles (HCAV). Journal of Advanced Transportation 17 5. Conclusion Tis study investigates the extent of confict severity resulting from the introduction of CAVs into trafc streams. It presents an analysis of the potential trafc conficts that occur when roads completely operating with HDVs transition into full L4 vehicle operation. Tree dimensions of severity are examined: proximity to collision, consequences of collision, and proximity/consequence of collision classifed by severity score. Owing to the lack of crash data involving CAVs, this study implemented a trafc microsimulation approach followed by SSAM analysis. Te specifc outputs of the SSAM (e.g., TTC, MaxS, DeltaS, and MaxDeltaV) are used to estimate confict severity. Te results of several mixed feet operation scenarios are compared to determine trafc safety when the real and current extent of penetration of CAVs on roads is exceeded. Te key fndings of this study are as follows. Te sensitivity analysis of the TTC threshold in scenarios where HCAV is the follower vehicle yields interesting results. If the presence of HCAV on the road is low (less than 35%), the diference in the number of identifed conficts between the applied TTC threshold values (i.e., 0.5, 0.75, 1.0, 1.25, and 1.5 s) is not statistically signifcant. By contrast, the scenarios where HCAVs have moderate sharing percentages (35%– 55%) start to show a signifcant diference at 1.0 s. Te scenarios where the operation percentage of these vehicles is high lead to signifcant diferences in the number of conficts among all the tested TTC values. Terefore, the importance of applying diferent TTC threshold values for such scenarios must be recognized. Te MaxS and DeltaS values were discussed as confict consequence indicators within the proposed scenarios and several vehicle interactions. Tese indicators show that the scenarios where 55% or more HCAV share the road result in conficts with low severity (low speeds and low speed diferences among vehicles involved in conficts). In addition, the conficts where HDVs are the follower vehicles yield the highest severity conficts, followed by the conficts where the follower vehicles are LCAV. Finally, proximity/consequence (TTC/MaxDeltaV) charts related to diferent vehicle types have been developed. Tese charts have been used to classify the resulting conficts into severity scores in each scenario. Te results indicate that increasing the shared percentages of CAVs operating on the road signifcantly decreases the number of conficts with high severity. When approximately 100% of HCAV operate on roads, severe conficts are anticipated to disappear, and those with low severity are reduced. Tis study presented a comprehensive investigation of trafc confict severity dimensions and analyzed the confict severity related to several levels of automation within various mixed feet operation scenarios. Nevertheless, this study presents some limitations that should be considered for future research. Firstly, whether the SSMs thresholds under conventional trafc conditions are applicable when modeling safety in mixed or fully automated trafc remains unclear. Diferent TTC threshold values have been tested and applied to solve this problem. However, when real data become available, the validity of SSM should be thoroughly reviewed and verifed. Terefore, new data sources related to CAV data will be crucial for the development of an universal SSM set that can satisfy all automation levels. Secondly, for particular trafc scenarios, the investigation of SSMs, such as the lateral safety provided by lane changing and merging maneuvers, must be implemented. Both HDVs and CAVs can exhibit diferent levels of lateral safety, particularly in a mixed autonomy trafc. And fnally, the calibration process could be improved with feld TTC data [63, 64]. Tis should be considered in similar future research. Appendix A. Behavior Parameters Used for CAV Levels Modeling Tis appendix contains the CAV behavior parameter values as indicated in Table 4 (for passenger cars) and Table 5 (for heavy vehicles). B. A Sample of Microsimulation Results Te following results in Figure 7 represent an example of the microsimulation outputs related to MaxS and DeltaS that were generated at scenario E, when HDV is the follower vehicle in the conficts. MaxS and DeltaS outputs are presented as distribution charts to refect the resulted data more descriptively. C. Severity Charts for HDVs and CAVs Tis appendix describes the procedure followed in developing the severity charts (by vehicle type) based on TTC score/MaxDeltaV score. To obtain the TTC score, diferent TTC thresholds were established by vehicle type. Te procedure looks for the infection points of the TTC cumulative distribution when pure vehicle type scenarios were modeled (i.e., all the vehicles in the simulation are exclusively HDVs, L1, L2, L3, or L4 vehicles). Specifcally, 15 microsimulation runs were executed for each pure scenario, and the TTC cumulative distribution charts were depicted. All the conficts identifed with a TTC value equal to or lower than 5.0 s were considered for the TTC distribution analysis. According to the Hyd´ en [21] safety pyramid, extremely severe conficts are considerably limited, whereas less severe trafc conficts are more frequent. According to Souleyrette and Hochstein [38], these severe conficts can be obtained based on the infection points of the TTC cumulative distribution of the pure scenarios for each vehicle type. Tese points are used as thresholds to delineate the few severe conficts from nonsevere ones. Later, the nonsevere conficts were divided into three approximately equal groups. A TTC score was assigned to each group (one severe and three nonsevere), which was later used to obtain the overall score. Table 6 summarizes the proposed TTC scores and thresholds to determine the overall scores of the pure operation scenarios. 18 Journal of Advanced Transportation As listed in Table 6, the thresholds that identify severe conficts (with a TTC score equal to 3) difer among the pure vehicle type operational scenarios. Precisely, the infection point for HCAV (L3 and L4 vehicles) was lower than those of the other vehicles, indicating their improved capabilities. Te infection points for the pure scenarios of HCAV, LCAV, and HDVs were 0.75, 1.0, and 1.5 s, respectively. Table 6 shows the variation in the infection point afects other thresholds. Moreover, the decrease in a severe confict region (with a score equal to 3) is clearly achieved by increasing the automation level, afrming the safety beneft of incorporating HCAV. For the consequence score, Souleyrette and Hochstein [38] used the equation of Evan [37], which employs MaxDeltaV to calculate the likelihood of crash injuries and fatalities. Teir results showed that MaxDeltaV values of approximately 30 and 60 km/h are key (infection) values that signifcantly increase the propensity for severe conficts. Because the Evan equation only depends on MaxDeltaV and the consequences of a crash with certain MaxDeltaV values have the same efect on HDVs and CAVs (with diferent automation levels), these two key values can be considered the same for all automation levels and interactions. Terefore, following Souleyrette and Hochstein [38], the severity value for MaxDeltaV is divided into three scores: (1) score 1, MaxDeltaV ranging 0–30 km/h; (2) score 2, MaxDeltaV ranging 30–60 km/h; and (3) score 3, MaxDeltaV exceeding 60 km/h. Te next step is adding both scores (TTC and MaxDeltaV scores) to obtain an overall severity score. Te resulted overall score is represented by regions in Figure 8(a), where each score area represents the severity score. However, severity scores are better identifed by lines or curves rather than by square areas [32, 38]. For this reason, these square areas are converted into severity isolines as contour lines (Figure 9). Precisely, each overall square area is reshaped to fve subscores (with an increment of 0.2 points) for each major score (in both ranges of TTC and MaxDeltaV scores), see Figure 8(b). Afterwards, the step-graded lines resulting from the equal overall scores in Figure 8(b) are reshaped into smooth contour lines for HDVs, LCAV, and HCAV, as shown in Figure 9. In the fgure, the variation of the red color from light to dark represents the increment in Severity Score (SS). Data Availability Te data supporting the current study are available from the corresponding author upon request. Disclosure Tis study is part of the Research Project PID2019110741RA-I00. Conflicts of Interest Te authors declare that there are no conficts of interest. Authors’ Contributions Roc´ ıo de Oña and Juan de Oña conceptualized the study; Tasneem Miqdady, Roc´ ıo de Oña, and Juan de Oña performed methodology; Tasneem Miqdady, Roc´ ıo de Oña, and Juan de Oña did formal analysis and investigation; Tasneem Miqdady wrote the original draft preparation; Roc´ ıo de Oña and Juan de Oña wrote the article and reviewed and edited the article; Roc´ ıo de Oña did funding acquisition. Acknowledgments Te authors are grateful to the Spanish General Directorate of Trafc (DGT) for providing the trafc fows from several GR-30 sections. Tasneem Miqdady appreciates Aimsun to provide their postgraduate student license to make this work. Tis work was supported by Research Project PID2019-110741RA-I00, fnanced by the Spanish State Research Agency (MCIN/AEI/10.13039/501100011033). Open Access funding enabled and organized by CRUE-CBUA Gold. References [1] Sae, Standard J3016: Taxonomy and Defnitions for Terms Related to On-Road Motor Vehicle Automated Driving Systems, 2014. [2] S. G. Shelby, “Delta-V as a measure of trafc confict severity,” Transportation Research Record, vol. 1, pp. 1–19, 2011. [3] T. Miqdady, R. DeOña, and J. DeOña, Quantifying the Safety Impact of Connected and Autonomous Vehicles in Motorways: A Simulation-Based Study, Libro de Actas), 2021. [4] M. M. Morando, Q. Tian, L. T. Truong, and H. L. Vu, “Studying the safety impact of autonomous vehicles using simulation-based surrogate safety measures,” Journal of Advanced Transportation, vol. 2018, Article ID 6135183, 11 pages, 2018. [5] A. Papadoulis, M. Quddus, and M. Imprialou, “Evaluating the safety impact of connected and autonomous vehicles on motorways,” Accident Analysis and Prevention, vol. 124, pp. 12–22, 2019. [6] A. Sinha, S. Chand, K. P. Wijayaratna, N. Virdi, and V. Dixit, “Comprehensive safety assessment in mixed feets with connected and automated vehicles: a crash severity and rate evaluation of conventional vehicles,” Accident Analysis and Prevention, vol. 142, Article ID 105567, 2020. [7] N. Virdi, H. Grzybowska, S. T. Waller, and V. Dixit, “A safety assessment of mixed feets with connected and autonomous vehicles using the surrogate safety assessment module,” Accident Analysis and Prevention, vol. 131, pp. 95–111, 2019. [8] M. Gu´ eriau and I. Dusparic, “Quantifying the impact of connected and autonomous vehicles on trafc efciency and safety in mixed trafc,” in Proceedings of the 2020 IEEE 23rd International Conference on Intelligent Transportation Systems, Rhodes, Greece, September 2020. [9] H. Xie, E. Tanin, S. Karunasekera et al., “Quantifying the impact of autonomous vehicles using microscopic simulations,” in Proceedings of the 12th International Workshop on Computational Transportation Science, Chicago, IL, USA, November 2019. [10] T. Miqdady, R. De Oña, J. Casas, and J. De Oña, “Studying trafc safety during the transition period between manual driving and autonomous driving: a simulation-based Journal of Advanced Transportation 19 approach,” IEEE Transactions on Intelligent Transportation Systems, p. 1, 2023. [11] L. Zheng, K. Ismail, and X. Meng, “Trafc confict techniques for road safety analysis: Open questions and some insights,” Canadian Journal of Civil Engineering, vol. 41, no. 7, pp. 633–641, 2014. [12] J. C. Hayward, “Near-miss determination through use of a scale of danger,” Highway Research Research Record, vol. 384, pp. 24–34, 1972. [13] L. Zheng and T. Sayed, “From univariate to bivariate extreme value models: approaches to integrate trafc confict indicators for crash estimation,” Transportation Research Part C: Emerging Technologies, vol. 103, pp. 211–225, 2019. [14] B. L. Allen, B. T. Shin, and P. J. Cooper, “Analysis of trafc conficts and collisions,” Transportation Research Record, vol. 667, pp. 67–74, 1978. [15] Aimsun, Aimsun Next 20 User´s Manual, Aimsun Next Version 20.0.2, Barcelona, Spain, 2020. [16] Y. Guo, T. Sayed, and M. Essa, “Real-time confict-based Bayesian Tobit models for safety evaluation of signalized intersections,” Accident Analysis and Prevention, vol. 144, Article ID 105660, 2020. [17] Y. Guo, T. Sayed, and L. Zheng, “A hierarchical bayesian peak over threshold approach for confict-based before-after safety evaluation of leading pedestrian intervals,” Accident Analysis and Prevention, vol. 147, Article ID 105772, 2020. [18] D. Gettman, L. Pu, T. Sayed, and S. Shelby, Surrogate Safety Assessment Model and Validation: Final Report, Publication No. FHWA-HRT-08-051, 2008. [19] G. R. Brow, “Trafc conficts for road user safety studies,” Canadian Journal of Civil Engineering, vol. 21, no. 1, pp. 1–15, 1994. [20] K. El-Basyouny and T. Sayed, “Safety performance functions using trafc conficts,” Safety Science, vol. 51, no. 1, pp. 160–164, 2013. [21] C. Hyd´en, Te Development of a Method for Trafc Safety Evaluation: Te Swedish Trafc Confict Technique Doctoral Tesis, Lund University, Department of Trafc Planning and Engineering, 1987. [22] D. Lord, “Analysis of pedestrian conficts with left-turning trafc,” Transportation Research Record, vol. 1538, 1996. [23] D. J. Migletz, W. D. Glauz, and K. M. Bauer, Relationships between Trafc Conficts and Accidents, U.S. Department of Transportation Federal Highway Administration, 1985. [24] E. Sacchi, T. Sayed, and P. de Leur, “A comparison of collision-based and confict-based safety evaluations: the case of right-turn smart channels,” Accident Analysis and Prevention, vol. 59, pp. 260–266, 2013. [25] J. Archer, Indicators for Trafc Safety Assessment and Prediction and Teir Applicaiton in Micro-simulation Modelling: A Study of Urban and Suburban IntersectionsDepartment of Infrastructure. Royal Institute of Technology, Stockholm, Sweden, 2005. [26] K. Ozbay, H. Yang, B. Bartin, and S. Mudigonda, “Derivation and validation of new simulation-based surrogate safety measure,” Transportation Research Record, vol. 2083, no. 1, pp. 105–113, 2008. [27] Federal Highway Administration (Fhwa), Surrogate Safety Assessment Model and Validation: Final Report, Federal Highway Administration, USA, 2008. [28] M. M. Minderhoud and P. H. L. Bovy, “Extended time-tocollision measures for road trafc safety assessment,” Accident Analysis and Prevention, vol. 33, no. 1, pp. 89–97, 2001. [29] D. F. Cooper and N. Ferguson, “Trafc studies at t-junctions - a confict simulation model,” Trafc Engineering and Control, vol. 17, pp. 306–309, 1976. [30] F. Cunto, Assessing Safety Performance of Transportation Systems Using Microscopic Simulation, Waterloo, Ontario, Canada, 2008. [31] C. Oh, S. Park, and S. G. Ritchie, “A method for identifying rear-end collision risks using inductive loop detectors,” Accident Analysis and Prevention, vol. 38, no. 2, pp. 295–301, 2006. [32] A. Laureshyn, T. De Ceunynck, C. Karlsson, A. Svensson, and S. Daniels, “In search of the severity dimension of trafc events: extended Delta-V as a trafc confict indicator,” Accident Analysis and Prevention, vol. 98, pp. 46–56, 2017. [33] A. Laureshyn, N. Saunier, and A. Fyhri, “Cross-comparison of three surrogate safety methods to diagnose cyclist safety problems at intersections in Norway,” Accident Analysis and Prevention, vol. 105, pp. 11–20, 2017. [34] W. K. M. Alhajyaseen, “Te integration of confict probability and severity for the safety assessment of intersections,” Arabian Journal for Science and Engineering, vol. 40, no. 2, pp. 421–430, 2015. [35] C. Wang, Y. Xie, H. Huang, and P. Liu, “A review of surrogate safety measures and their applications in connected and automated vehicles safety modeling,” Accident Analysis and Prevention, vol. 157, Article ID 106157, 2021. [36] W. L. Carlson, “Crash injury prediction model,” Accident Analysis and Prevention, vol. 11, no. 2, pp. 137–153, 1979. [37] L. Evans, “Driver injury and fatality risk in two-car crashes versus mass ratio inferred using Newtonian mechanics,” Accident Analysis and Prevention, vol. 26, no. 5, pp. 609–616, 1994. [38] R. R. Souleyrette and J. L. Hochstein, Development of a Confict Analysis Methodology Using SSAM, 2012. [39] R. van der Horst and R. J. Kraay, Te Dutch Confict Technique −DOCTOR, ICTCT Workshop, Budapest, 1986. [40] P. G˚ arder, “Konfiktstudier i landsv¨ agskorsningar (in Swedish) Confict studies in rural intersections,” Bulletin, vol. 42, Lund, Institutionen f¨ or Trafkteknik, 1982. [41] L. Shbeeb, Development of a Trafc Conficts Technique for Diferent Environments - a Comparative Study of Pedestrian Conficts in Sweden and Jordan, Doctoral thesis,University of Lund, Lund Institute of Technology, Department of Technology and Society, Trafc Engineering, 2000. [42] A. Svensson, “A method for analysing the trafc process in a safety perspective,” Bulletin, Lund Institute of Technology, Lund University, vol. 166, 2010 [43] Y. El-Hansali, S. Farrag, A. Yasar, E. Shakshuki, and K. AlAbri, “Using surrogate measures to evaluate the safety of autonomous vehicles,” Procedia Computer Science, vol. 191, pp. 151–159, 2021. [44] M. S. Rahman, M. Abdel-Aty, J. Lee, and M. H. Rahman, “Safety benefts of arterials’ crash risk under connected and automated vehicles,” Transportation Research Part C: Emerging Technologies, vol. 100, pp. 354–371, 2019. [45] J. Zhang, K. Wu, M. Cheng, M. Yang, Y. Cheng, and S. Li, “Safety evaluation for connected and autonomous vehicles’ exclusive lanes considering penetrate ratios and impact of trucks using surrogate safety measures,” Journal of Advanced Transportation, vol. 2020, Article ID 5847814, 16 pages, 2020. [46] A. Sinha, V. Vu, S. Chand, K. Wijayaratna, and V. Dixit, “A crash injury model involving autonomous vehicle: investigating of crash and disengagement reports,” Sustainability, vol. 13, no. 14, p. 7938, 2021. 20 Journal of Advanced Transportation [47] H. Chen, H. Chen, Z. Liu, X. Sun, and R. Zhou, “Analysis of factors afecting the severity of automated vehicle crashes using XGBoost model combining POI data,” Journal of Advanced Transportation, vol. 2020, pp. 1–12, 2020. [48] Y. Li, Z. Li, H. Wang, W. Wang, and L. Xing, “Evaluating the safety impact of adaptive cruise control in trafc oscillations on freeways,” Accident Analysis and Prevention, vol. 104, pp. 137–145, 2017. [49] Z. Zhong, J. Lee, and L. Zhao, “Trafc fow characteristics and lane use strategies for connected and automated vehicles in mixed trafc conditions,” Journal of Advanced Transportation, 2021. [50] Y. Li, Y. Tu, Q. Fan, C. Dong, and W. Wang, “Infuence of cyber-attacks on longitudinal safety of connected and automated vehicles,” Accident Analysis and Prevention, vol. 121, pp. 148–156, 2018. [51] M. S. Rahman and M. Abdel-Aty, “Longitudinal safety evaluation of connected vehicles’ platooning on expressways,” Accident Analysis and Prevention, vol. 117, pp. 381–391, 2018. [52] Y. Tu, W. Wang, Y. Li, C. C. Xu, T. Xu, and X. Li, “Longitudinal safety impacts of cooperative adaptive cruise control vehicle’s degradation,” Journal of Safety Research, vol. 69, pp. 177–192, 2019. [53] A. D. Tibljaˇ s and T. Giufr, “Introduction of autonomous vehicles: roundabouts design and safety performance evaluation,” Sustainability, vol. 10, no. 1060, pp. 1–14, 2018. [54] W. Weijermars, A. Hula, A. Chaudhry et al., “LEVITATE: road safety impacts of connected and automated vehicles,” 2021, https://levitate-project.eu/downloads/. [55] K. Wunderlich, M. Vasudevan, and P. Wang, Trafc Analysis toolbox volume III: Guidelines for applying trafc microsimulation modeling software 2019 update to the 2004 version, 2019. [56] Roads and Maritime Services, Trafc Modelling Guidelines, Roads and Maritime Services, Sydney, 2013. [57] P. G. Gipps, “A behavioural car-following model for computer simulation,” Transportation Research Part B: Methodological, vol. 15, no. 2, pp. 105–111, 1981. [58] P. G. Gipps, “A model for the structure of lane-changing decisions,” Transportation Research Part B: Methodological, vol. 20, no. 5, pp. 403–414, 1986. [59] Atkins, Research on the Impacts of Connected and Autonomous Vehicles (CAVs) on Trafc Flow Summary Report Department for Transport, 2016. [60] S. Das and A. K. Maurya, “Defning time-to-collision thresholds by the type of lead vehicle in non-lane-based trafc environments,” IEEE Transactions on Intelligent Transportation Systems, vol. 21, no. 12, pp. 4972–4982, 2020. [61] E. Papazikou, M. Zach, H. C. Boghani et al., “Detailed list of sub-use cases, applicable forecasting methodologies and necessary output variables,” Deliverable D4.4 of the H2020 Project LEVITATE, 2020. [62] D. Gettman and L. Head, “Surrogate safety measures from trafc simulation models,” Report No. FHWA-RD-03-050, 2003. [63] Y. Guo, M. Essa, T. Sayed, M. M. Haque, and S. Washington, “A comparison between simulated and feld-measured conficts for safety assessment of signalized intersections in Australia,” Transportation Research Part C: Emerging Technologies, vol. 101, pp. 96–110, 2019. [64] Y. Guo, T. Sayed, L. Zheng, and M. Essa, “An extreme value theory based approach for calibration of microsimulation models for safety analysis,” Simulation Modelling Practice and Teory, vol. 106, Article ID 102172, 2021. [65] G. Mesionis, M. Brackstone, and N. Gravett, “Microscopic modeling of the efects of autonomous vehicles on motorway performance,” Transportation Research Record, vol. 2674, no. 11, pp. 697–707, 2020. [66] L. Ye and T. Yamamoto, “Evaluating the impact of connected and autonomous vehicles on trafc safety,” Physica A: Statistical Mechanics and Its Applications, vol. 526, Article ID 121009, 2019. [67] D. Stanek, E. Huang, R. Milam, and A. Wang, Measuring Autonomous Vehicle Impacts on Congested Networks Using Simulation, 2017. [68] J. Karjanto, N. Yusof, J. Terken, and F. Delbressine, Simulating Autonomous Driving Styles: Accelerations for Tree Road Profles, vol. 1–16, 2017. [69] A. K. Bakhshi and M. M. Ahmed, Accounting for HumanRelated Unobserved Heterogeneity in the Safety Performance of Connected Vehicles: An Incorporation of Bayesian Hierarchical Negative Binomial into Simulated Work Zone Warning Application, IATSS Research, 2021. Journal of Advanced Transportation 21