Accident; analysis and prevention最新文献

{"title":"S2O: An Integrated driving decision-making performance evaluation method bridging subjective feeling to objective evaluation","authors":"Yuning Wang ,&nbsp;Zehong Ke ,&nbsp;Yanbo Jiang ,&nbsp;Jinhao Li ,&nbsp;Shaobing Xu ,&nbsp;John M. Dolan ,&nbsp;Jianqiang Wang","doi":"10.1016/j.aap.2025.108026","DOIUrl":"10.1016/j.aap.2025.108026","url":null,"abstract":"<div><div>Autonomous driving decision-making is one of the critical modules towards intelligent transportation systems, and how to evaluate the driving performance comprehensively and precisely is a crucial challenge. A biased evaluation misleads and hinders decision-making modification and development. Current planning evaluation metrics include deviation from the real driver trajectory and objective driving experience indicators. The former category does not necessarily indicate good driving performance since human drivers also make errors and has been proven to be ineffective in interactive close-loop systems. On the other hand, existing objective driving experience models only consider limited factors, lacking comprehensiveness. And the integration mechanism of various factors relies on intuitive experience, lacking precision. In this research, we propose <span>S2O</span>, a novel integrated decision-making evaluation method bridging subjective human feeling to objective evaluation. First, modified fundamental models of four kinds of driving factors which are safety, time efficiency, comfort, and energy efficiency are established to cover common driving factors. Then based on the analysis of human rating distribution regularity, a segmental linear fitting model in conjunction with a complementary SVM segment classifier is designed to express human’s subjective rating by objective driving factor terms. Experiments are conducted on the D2E dataset, which includes approximately 1,000 driving cases and 40,000 human rating scores. Results show that <span>S2O</span> achieves a mean absolute error of 4.58 to ground truth under a percentage scale. Compared with baselines, the evaluation error is reduced by 32.55%. Implementation on the SUMO platform proves the real-time efficiency of online evaluation, and validation on performance evaluation of three autonomous driving planning algorithms proves the feasibility.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"217 ","pages":"Article 108026"},"PeriodicalIF":5.7,"publicationDate":"2025-04-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143830117","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Exploring the inter-relationship among hazardous actions, causal factors, and motorcyclist injury severity using path analysis","authors":"Qiong Yu ,&nbsp;Jiageng Niu ,&nbsp;Jushang Ou ,&nbsp;Wei Bai ,&nbsp;Nengfeng Wang ,&nbsp;Xinguo Jiang","doi":"10.1016/j.aap.2025.108033","DOIUrl":"10.1016/j.aap.2025.108033","url":null,"abstract":"<div><div>Hazardous actions are the critical contributor to motorcyclist injury severity, which are typically treated in parallel with other causal factors in crash severity models. However, hazardous actions may result directly from the interaction between road users, roadway, and environmental factors. So, causal factors may indirectly influence motorcyclist injury severity through hazardous actions. The research attempts to investigate the direct, indirect, and total effects of causal factors on motorcyclist injury severity, emphasizing the role of hazardous actions as mediators. A path analysis approach is employed to model the complex relationship among hazardous actions, other causal factors, and motorcyclist injury severity. Specifically, two models are developed within the path analysis framework, namely, a random parameters logit model to examine the relationship between contributing factors and hazardous actions, and a correlated random parameters logit model with heterogeneity in means to assess the correlation between these factors and motorcyclist injury severity. Using motorcycle crash data in Michigan from 2015 to 2018, the study finds that factors such as stop/yield signs, intersections, one-way traffic, two-lane roads, midnight/early morning, weekends, and dawn/dusk do not directly affect motorcyclist injury severity but significantly influence hazardous actions, thereby indirectly increasing injury severity through these actions. Moreover, alcohol or drug use, speed limits of 55 mph or higher, and motorcycle age of six years or older significantly increase the risk of hazardous actions and motorcyclist injury severity, thus exacerbating injury severity through the indirect influence of hazardous actions. Additionally, divided median strips with or without traffic barriers, rain/snow weather, and darkness with streetlights directly reduce injury severity but indirectly increase it via hazardous actions. These findings underscore the complex inter-relationship among hazardous actions, other causal factors, and motorcyclist injury severity, offering valuable insights for enhancing motorcycle safety by addressing hazardous actions as a central factor.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"216 ","pages":"Article 108033"},"PeriodicalIF":5.7,"publicationDate":"2025-04-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143823896","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Children on wheels: Identifying crash determinants using cluster correspondence analysis","authors":"Rohit Chakraborty,&nbsp;David Mills,&nbsp;Subasish Das","doi":"10.1016/j.aap.2025.108025","DOIUrl":"10.1016/j.aap.2025.108025","url":null,"abstract":"<div><div>Child bicyclists (14 years old and younger) are among the most vulnerable road users, facing significant risks of crashes that often result in severe injuries or fatalities. This study aims to identify key factors influencing child bicyclist crashes and uncover distinct crash patterns using a dataset of 2,394 crashes in Texas from 2017 to 2022. Employing a hybrid approach through machine learning models XGBoost and Random Forest, and Cluster Correspondence Analysis (CCA), the research identified six clusters characterized by unique crash factors and patterns. Moreover, SHAP analysis was conducted on each cluster to further investigate the impact of factors on crash severity. Intersection-related crashes were driven by driver behavior at stop signs and signalized intersections, while urban crashes highlighted risks at marked lanes and driveway access areas. Crashes in residential and rural areas revealed vulnerabilities due to limited traffic control and infrastructure, with rural areas further increased by higher vehicle speeds. Crashes under adverse weather conditions and poor lighting emphasized the role of environmental hazards in increasing crash risks. This study provides countermeasures, including intersection redesigns with protected crossings, expansion of separated bike lanes, and enhanced driveway management. Improved lighting, high-friction road coatings, and weather-specific safety campaigns are recommended to address environmental risks. From a policy perspective, the findings highlight the need for equitable infrastructure investments, stricter enforcement of traffic laws, and educational programs targeting both child bicyclists and drivers. This study also developed an interactive visualization tool that allows users to explore crash locations and related factors. By addressing these challenges, this research offers a framework for improving child bicyclist safety and advancing safer road environments.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"216 ","pages":""},"PeriodicalIF":5.7,"publicationDate":"2025-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143814969","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Understanding motorcycle crash involvement: Insights from regular motorcyclists and food delivery riders in Vietnam","authors":"Amjad Pervez ,&nbsp;Duy Q. Nguyen-Phuoc ,&nbsp;Nhat Xuan Mai ,&nbsp;Dinh Quang Nhat Vo ,&nbsp;Jaeyoung Jay Lee","doi":"10.1016/j.aap.2025.108024","DOIUrl":"10.1016/j.aap.2025.108024","url":null,"abstract":"<div><div>Motorcycles have become a primary mode of transportation in many low- and middle-income countries, including Vietnam, where they are widely used for personal transport and commercial activities. The growing reliance on motorcycles, driven by rapid urbanization and the rise of app-based delivery platforms, has brought economic benefits but also significant public health concerns due to the high incidence of road traffic crashes. This study, based on a questionnaire survey targeting regular motorcyclists and food delivery riders in Vietnam, examines and compares the factors contributing to crash involvement between two groups, regular and delivery riders. Random parameters models with heterogeneity in means and variances were employed to capture variability in respondent behaviors. The results reveal that psychological factors, such as negative attitudes toward traffic rules and intentions to violate rules, significantly increase crash involvement for both regular and delivery riders, with the effect being more pronounced among delivery riders due to job pressures. Conversely, positive attitudes reduce crash involvement but are less effective for delivery riders due to high time pressures and frequent distractions. Perceived severity of crashes and the swiftness of sanctions also play critical roles: higher perceived severity promotes safer behavior, while the threat of sanctions deters risky actions for both groups. Rider attributes, such as age, education, and income, influence crash involvement, with younger and lower-educated riders facing higher risks among both groups due to inexperience and limited traffic knowledge, while income effects vary between regular and delivery riders. Travel characteristics, such as travel durations of more than two hours, increase crash involvement due to fatigue and exposure, particularly affecting delivery riders who navigate complex urban environments and face frequent distractions. The findings also highlight the importance of addressing heterogeneity in data analysis for more comprehensive insights. Moreover, based on these results, various policy implications are provided to reduce traffic crashes and enhance safety for motorcyclists in motorcycle-dominated countries.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"216 ","pages":""},"PeriodicalIF":5.7,"publicationDate":"2025-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143807107","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Examining the impact of centerline rumble strips on reducing rural two-lane head-on collisions in Maine","authors":"Jhan Kevin Gil-Marin ,&nbsp;Alainie Sawtelle ,&nbsp;Per Erik Garder ,&nbsp;Mohammadali Shirazi","doi":"10.1016/j.aap.2025.108003","DOIUrl":"10.1016/j.aap.2025.108003","url":null,"abstract":"<div><div>Among all traffic collisions, lane departure crashes are the leading type of serious traffic crashes in Maine, comprising 73% of statewide traffic fatalities. To reduce these crashes, the Maine Department of Transportation (MaineDOT) installed shoulder and centerline rumble strips on roadways to prevent lane departure crashes. Specifically, 511 miles of centerline rumble strips were installed on undivided bidirectional rural two-lane roadways to prevent head-on collisions. Given the severity of head-on collisions, coupled with significant investment in rumble strip installation, there is a need to understand the impact of rumble strips in reducing lane departure crashes. This study uses observational before-and-after studies with two methods: comparison group, and empirical Bayes (EB) comparison group to explore the effectiveness of centerline rumble strips in reducing head-on and opposite sideswipe crashes on rural two-lane roadways and compute crash modification factors (CMFs) in Maine. The evaluation investigated the impact of centerline rumble strips on reducing both the total as well as fatal and injury head-on and sideswipe collisions on rural two-lane roadways. The economic benefits of using rumble strips are also explored using a benefit-cost ratio analysis. This study finds that the installation of centerline rumble strips is associated with reductions of 28%-48% of head-on and opposite sideswipe collisions on rural two-lane roads. In addition, the benefits of the rumble strip installations are at least 14 times the cost. The centerline rumble strips are cost-effective countermeasures to reduce head-on collisions on rural two-lane roadways in Maine.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"216 ","pages":"Article 108003"},"PeriodicalIF":5.7,"publicationDate":"2025-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143790804","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Exploring how physio-psychological states affect drivers’ takeover performance in conditional automated vehicles","authors":"Ange Wang ,&nbsp;Jiyao Wang ,&nbsp;Chunxi Huang ,&nbsp;Dengbo He ,&nbsp;Hai Yang","doi":"10.1016/j.aap.2025.108022","DOIUrl":"10.1016/j.aap.2025.108022","url":null,"abstract":"<div><div>Although driving automation is promised to improve driving safety, drivers are still required to take over the control of the vehicles in case of emergency. Estimating drivers’ takeover performance serves as the basis for adaptive driving automation and takeover request (TOR) to ensure driving safety. However, although algorithms have been proposed to estimate drivers’ takeover performance through physiological and eye-tracking measures, the complex interrelationships between these metrics and driver behavior, as well as the interactions among the metrics themselves, are not fully understood. To answer this question, a driving simulation experiment involving 42 participants was conducted. Drivers experienced three types of takeover scenarios requested by TOR while driving a conditionally automated vehicle. Drivers’ physiological, eye-tracking metrics and psychological states, as imposed by several non-driving-related tasks were collected. A structural equation model was used to explore the interactions among physiological metrics (i.e., cardiac activity, respiratory activity, electrodermal activity), eye-tracking metrics, psychological states (i.e., trust in driving automation and perceived workload), and variations in takeover time and takeover quality. The results showed that trust was positively associated with takeover quality, while workload was positively associated with takeover time. Additionally, physiological and eye-tracking metrics were indirectly associated with takeover quality via psychological states. This study reveals the hierarchical relationship among takeover-performance-related variables and provides insights for designing driver monitoring systems aimed at estimating takeover performance in vehicles with driving automation and adaptive driving automation to improve driving safety.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"216 ","pages":""},"PeriodicalIF":5.7,"publicationDate":"2025-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143785673","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Reliable imputation of incomplete crash data for predicting driver injury severity","authors":"Xiaowei Gao ,&nbsp;Xinke Jiang ,&nbsp;Dingyi Zhuang ,&nbsp;James Haworth ,&nbsp;Shenhao Wang ,&nbsp;Ilya Ilyankou ,&nbsp;Huanfa Chen","doi":"10.1016/j.aap.2025.108020","DOIUrl":"10.1016/j.aap.2025.108020","url":null,"abstract":"<div><div>Traffic crash analyses are frequently challenged by incomplete documentation, particularly in standardised multi-party crash full records. Traditional imputation methods like MICE and KNN, while effective for single-category analyses, fail to address the complex interdependencies inherent in standardised crash records where different types of road user are present. This study introduces a novel graph-based imputation framework that integrates an Inexact Match Bipartite-Graph with Contrastive Learning in a Transformer-GNN architecture, providing a unified solution to handle missing data of various crash types in a complete crash record database. Testing on UK traffic crash records (2018–2022) demonstrates the robust performance of the imputation model, achieving imputation accuracy between 99.24% and 94.74% across missing data rates from 10% to 70%. In the downstream task of classifying the severity of the injury, our imputed data set proved to be highly reliable, achieving a Gmean score of 62.19% to identify levels of imbalanced severity, even under severe missing with a missing rate of 70%. Furthermore, explainable SHAP values demonstrated that data imputation preserved the most important contributing factors. These results validate our framework’s effectiveness in maintaining both data integrity and essential relationship structures in standardised crash records, advancing the field of traffic safety analysis through improved imputation methodology.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"216 ","pages":"Article 108020"},"PeriodicalIF":5.7,"publicationDate":"2025-04-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143783838","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Exploring the effect of driver drowsiness on takeover performance during automated driving: An updated literature review","authors":"Hengyan Pan ,&nbsp;David B. Logan ,&nbsp;Amanda N. Stephens ,&nbsp;William Payre ,&nbsp;Yonggang Wang ,&nbsp;Zhipeng Peng ,&nbsp;Yang Qin ,&nbsp;Sjaan Koppel","doi":"10.1016/j.aap.2025.108023","DOIUrl":"10.1016/j.aap.2025.108023","url":null,"abstract":"&lt;div&gt;&lt;h3&gt;Introduction&lt;/h3&gt;&lt;div&gt;Vehicle automation technology has considerable potential for reducing road crashes associated with human error, including issues related to driver drowsiness. However, before full automation becomes available on public roads, it will be essential for drivers to take back control from automated driving systems when requested. This poses a challenge for drivers, particularly as automation may further exacerbate drowsiness. This paper aims to update a systematic review published in 2022 (Merlhiot &amp; Bueno, Accident Analysis and Prevention, 170, 106536), to discuss factors affecting driving drowsiness and takeover performance with a particular focus on those not identified in previous review.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Method&lt;/h3&gt;&lt;div&gt;Following the Preferred Reporting Items for Systematic Reviews and Meta-analyses guidelines, three databases: &lt;em&gt;Web of Science&lt;/em&gt;, &lt;em&gt;PubMed&lt;/em&gt; and &lt;em&gt;Scopus&lt;/em&gt; were searched for studies published between March 2021 and October 2024. The following eligibility criteria were applied for study inclusion: 1) participants must have interacted with a simulated or real-world vehicle featured with driving automation Level 2 or above; 2) with at least one measurement indicator of driver drowsiness; 3) with at least one measurement indicator of takeover performance; 4) be conducted within a controlled experimental design. From an initial selection of 182 articles from databases, a total of twelve published articles were obtained after removing duplicates, title, abstracts and full texts checking. Additionally, 17 articles from the previous review were included, resulting in a total of 29 articles for this review study.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Results&lt;/h3&gt;&lt;div&gt;Driver drowsiness (e.g, increased Karolinska Sleepiness Scale levels, blink frequency) tended to increase with both the duration of automated driving and automation levels. Engaging in non-driving related tasks (NDRTs) alleviates drowsiness (e.g, lower heart rate and percentage of eye closure), but reduces takeover performance (e.g., longer braking reaction times, stronger longitudinal acceleration, shorter minimal time to collision). Compared to older drivers, younger drivers were more susceptible to drowsiness, while older drivers had worse takeover performance (e.g., delayed steering reaction time, higher collision rates). Sleep inertia and circadian rhythms were also identified as factors influencing takeover performance. The road monitoring task helps prevent excessive participation in NDRTs and improves takeover performance (e.g, reduced brake reaction times and maximum steering velocity, increased the minimum time to collision). Digital voice assistants and scheduled manual driving help maintain alertness (e.g, decreased blink duration) and enhance takeover performance (e.g, shorter reaction time to resume steering).&lt;/div&gt;&lt;div&gt;There were several limitations of the methodologies applied in the existing studies, among which were: 1) a la","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"216 ","pages":"Article 108023"},"PeriodicalIF":5.7,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143739227","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An integrated potential safety hazard assessment framework in connected car-following scenario","authors":"Guilong Xu ,&nbsp;Zhen Yang ,&nbsp;Jinfeng Ying ,&nbsp;Shikun Xie ,&nbsp;Shumin Bai ,&nbsp;Yani Qi","doi":"10.1016/j.aap.2025.108010","DOIUrl":"10.1016/j.aap.2025.108010","url":null,"abstract":"<div><div>In continuous car-following scenarios, a minor conflict could be amplified along the platoon over time due to system instability, resulting in high-risk rear-end situations. Conventional surrogate safety measures (SSM) only adopt the motion information of ego vehicle and preceding one, failing to capture potential safety hazard beyond the leading vehicle in the platoon. Identifying potential safety hazard beyond the 1st preceding one is substantially important for connected and automated vehicles (CAVs) which could take proactive actions to prevent high-risk scenarios. Toward this end, drawing on reliability theory, the current paper developed an integrated potential safety hazard assessment framework considering the motion information of preceding multiple vehicles. An innovative surrogate safety measures (SSM), named the potential safety hazard index (PSHI), is developed to capture potential safety hazard from preceding multiple vehicles in the car-following environment. We tested the validity of proposed PSHI in a real-world rear-end crash scenario. To real-timely apply PSHI to CAVs, we developed an orthogonal transformation first order reliability method (OTFORM) to accelerate computational time, keeping computation burden within 0.01 s. Another worthwhile finding is potential safety hazard mainly comes from preceding three vehicles in risky car-following cases. The new potential safety hazard evaluation framework provides a creative perspective for safety assessment and also shows a good prospect for longitudinal safety control of CAVs strategies.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"216 ","pages":"Article 108010"},"PeriodicalIF":5.7,"publicationDate":"2025-03-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143739226","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Joint analysis of crash injury severities for autonomous and conventional vehicles in mixed traffic environments: Application of random parameter bivariate probit model","authors":"Jian Xiang ,&nbsp;Zhengwu Wang ,&nbsp;Yibo Chen ,&nbsp;Ziran Meng ,&nbsp;Jie Wang","doi":"10.1016/j.aap.2025.108017","DOIUrl":"10.1016/j.aap.2025.108017","url":null,"abstract":"<div><div>Autonomous vehicles (AVs) are expected to significantly enhance road safety in the future. However, until fully autonomous driving systems are widely adopted, mixed traffic with AVs and conventional vehicles (CVs) will remain a typical feature of roadways. Consequently, it is crucial to understand how roadway and built environment factors impact traffic safety in mixed traffic settings. This study proposes a joint model to analyze crash injury severity for both autonomous and conventional vehicles within a unified framework. A random parameter bivariate probit model (RBP) is used as the methodological approach, as it accounts for the correlation between injury outcomes for AVs and CVs, while also capturing unobserved heterogeneity among the factors influencing safety. The model is developed using a dataset of 699 paired crashes, involving both AVs and CVs, occurring in proximity to each other in mixed traffic conditions in California. For comparison, both a random parameters univariate probit model (RUP) and a bivariate probit model (BP) are also developed. Model comparison results demonstrate that the proposed RBP model outperforms both the RUP and BP model in terms of explanatory power and goodness-of-fit. The parameter estimates reveal divergent effects of crash type and cause, natural environmental conditions, roadway features, and built environment factors on injury severity for autonomous and conventional vehicle crashes. The key results include: (1) A primary cause of AV crashes is the failure of CV drivers to respond appropriately or in a timely manner to unexpected changes in AV behaviors. (2) Adverse natural conditions, such as dark, pose a greater safety risk for AVs compared to CVs. (3) Road features with complex traffic conditions—such as Y-shaped intersections, traffic signals, and areas where lanes merge or diverge—are associated with a higher likelihood of injury in AV crashes, whereas these factors do not significantly affect injury severity in CV crashes. (4) Built environment factors related to vulnerable road users and public transportation infrastructure, such as crosswalks, schools, bus stops, and metro stops, exhibit notably heterogeneous effects on injury severity in AV crashes. The findings of this study have important implications for developing targeted strategies to enhance safety in mixed traffic environments. These strategies include establishing effective communication systems between autonomous and conventional vehicles, improving obstacle detection and performance in low-visibility conditions, and ensuring well-equipped road infrastructure for vulnerable road users.</div></div>","PeriodicalId":6926,"journal":{"name":"Accident; analysis and prevention","volume":"215 ","pages":"Article 108017"},"PeriodicalIF":5.7,"publicationDate":"2025-03-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143734565","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}