Ambiguous Driving Scenarios Survey

Abstract

The Ambiguous Driving Scenarios Survey (Baby et al., 2024) is a specialized psychometric and human-factors assessment instrument designed to categorize, evaluate, and quantify human perceptions of ambiguous driving scenarios (ADS) encountered by autonomous vehicles (AVs). In transportation psychology and automated engineering, an ambiguous driving scenario refers to an operational road environment in which an automated system cannot reliably infer the trajectory, intentions, or behavioral state of other road users due to environmental occlusion, unstructured social signaling, or sensor equivocation. Developed through a multi-stage methodology involving the synthesis of over 10,000 empirical scenarios, an expert-driven filtering to 6,590 situations, and user experience (UX) card sorting protocols, the final instrument operationalizes 28 representative ambiguous driving scenarios. Each scenario is evaluated across five latent dimensions of user evaluation: Moral Perception, Ethical Perception, Utility Perception, Legal Perception, and Safety Perception. Evaluated using a five-point Likert response format across an adult cohort of drivers in the Republic of Korea, the instrument demonstrates robust psychometric properties. Structural integrity was confirmed via confirmatory factor analysis (CFA), demonstrating strong construct validity. Convergent validity is evidenced by average variance extracted (AVE) estimates exceeding 0.50 and standardized factor loadings surpassing 0.60. Discriminant validity satisfies the Fornell-Larcker criterion, with the square root of each factor’s AVE exceeding its inter-construct correlations. Composite reliability coefficients range from 0.800 to 0.851, and split-half reliability coefficients exceed 0.812 across all subscales. The survey serves as an empirical foundation for human-machine interface (HMI) engineers, automotive software architects, and policymakers tasked with harmonizing AV trajectory planning algorithms with human socio-cognitive expectations.

Keywords

Automated Driving Systems, Autonomous Vehicles, Ambiguous Driving Scenarios, Moral Perception, Ethical Perception, Legal Perception, Utility Perception, Safety Perception, Human Factors Engineering, Traffic Psychology

Authors

The Ambiguous Driving Scenarios Survey was developed by an international, interdisciplinary consortium of researchers spanning human factors engineering, computer science, digital media, and transportation safety:

• Tiju Baby — Division of Media, Culture, and Design Technology, Hanyang University ERICA, Republic of Korea (ORCID).
• Hatice Şahin Ippoliti — Department of Computing Science, University of Oldenburg, Germany.
• Philipp Wintersberger — Digital Media Department, University of Applied Sciences Upper Austria, Hagenberg, Austria (ORCID).
• Yiqi Zhang — Department of Industrial and Manufacturing Engineering, Pennsylvania State University, University Park, PA, United States.
• Sol Hee Yoon — Department of Safety Engineering, Seoul National University of Science and Technology, Seoul, Republic of Korea.
• Jieun Lee — Department of Safety Engineering, Pukyong National University, Busan, Republic of Korea.
• Seul Chan Lee (Corresponding Author) — Division of Media, Culture, and Design Technology, Hanyang University ERICA, Lion’s Hall 202, 55 Hanyangdaehak-ro, Sangnok-gu, Ansan-si, Gyeonggi-do, Republic of Korea. Email: [email protected].

Purpose

The primary purpose of the Ambiguous Driving Scenarios Survey is to systematically bridge the gap between algorithmic decision-making in automated driving systems (ADS) and the socio-cognitive, normative expectations held by human drivers and society. As vehicles transition toward higher levels of driving automation (e.g., SAE International Levels 3, 4, and 5), automated systems must operate not merely in deterministic physical environments, but within dynamic, socially negotiated traffic ecosystems. In everyday driving, human road users continuously navigate ambiguous edge cases—such as an unbelted pedestrian hesitating at an unmarked midblock crossing, an oncoming vehicle partially intruding over a double-yellow line to avoid debris, or a cyclist signaling ambiguous intentions via gaze shifts. While human drivers resolve these dilemmas through implicit social heuristics, defensive driving, and intuitive risk appraisals, autonomous vehicles struggle with programmatic hesitation, overly conservative freezes, or potentially hazardous maneuvers.

Consequently, the instrument addresses several critical operational and theoretical challenges:

• Algorithmic Behavioral Calibration: AV motion-planning algorithms cannot optimize for single-variable objectives (such as purely minimizing travel time or strictly obeying speed limits) without generating unacceptable risks or societal backlash. The survey quantifies the relative importance of distinct normative dimensions—safety, legality, moral obligation, ethical harm minimization, and utility—allowing control system engineers to weight multi-objective cost functions according to human consensus.
• Classification of Ambiguous Edge Cases: Prior research in automotive safety predominantly targeted either unambiguous baseline driving (e.g., standard lane keeping, adaptive cruise control) or catastrophic, forced-choice life-and-death dilemmas (e.g., classic philosophical trolley problems). The survey provides a standardized taxonomy of non-binary, highly frequent ambiguous operational design domain (ODD) scenarios that occur in naturalistic urban and suburban transit.
• Policy and Regulatory Harmonization: Regulatory agencies require empirical frameworks to evaluate AV safety cases. By demonstrating how human road users prioritize legality versus utility or safety versus strict regulatory adherence (e.g., crossing a solid line to provide safe clearance for a vulnerable road user), the instrument supports the drafting of traffic laws adapted to autonomous agents.
• Human-Machine Trust and Acceptance: A primary barrier to the consumer adoption of automated mobility is the perceived unpredictability or alien driving style of AVs. Administering this instrument enables human factors researchers to identify which contextual scenario attributes trigger moral outrage, perceived illegitimacy, or user discomfort, guiding the development of transparent human-machine interfaces (HMIs).

Psychological Construct

The Ambiguous Driving Scenarios Survey measures human cognitive, evaluative, and normative perceptions regarding autonomous vehicle responses in driving scenarios characterized by high epistemic uncertainty. The overarching construct—Perception of Ambiguous Driving Scenarios—is defined as a multidimensional evaluative framework through which individuals assess the appropriateness, urgency, and moral permissibility of an automated vehicle’s actions when interacting with ambiguous environmental stimuli. The construct is decomposed into five distinct, interrelated sub-dimensions:

1. Moral Perception

Moral Perception encompasses an individual’s intuitive recognition and appraisal of fundamental rightness, wrongness, and personal virtue within a driving event. Unlike formal legal codes, moral evaluations in traffic stem from shared human intuitions regarding care, vulnerability, and blameworthiness. In the context of ambiguous driving, moral perception is activated when an AV’s trajectory involves potential psychological distress, implicit social commitments, or disproportionate risks imposed upon vulnerable entities. For example, in Scenario 5 and Scenario 11, where an automated vehicle encounters ambiguous signals from a pedestrian near a crosswalk, respondents evaluate whether the vehicle exhibits basic humanitarian care and respect for human vulnerability, independent of strict right-of-way ordinances.

2. Ethical Perception

Ethical Perception reflects structured normative reasoning concerning justice, equity, rights, and the principled distribution of risk across multiple stakeholders. While moral perception is often rooted in visceral social intuitions, ethical perception operationalizes deontological and teleological frameworks to assess fairness. Within scenarios such as Scenario 1, 4, 17, and 23, ethical perception evaluates whether the AV distributes physical vulnerability equitably (e.g., between an erratic micromobility user and vehicle occupants) or whether it unfairly burdens certain road users to preserve its own efficiency.

3. Utility Perception

Utility Perception reflects an evaluation rooted in utilitarianism and systemic efficiency, focusing on the maximization of overall collective benefit and the minimization of operational waste (e.g., traffic delays, energy consumption, bottlenecks). In ambiguous traffic contexts, an overly cautious AV that comes to a complete standstill whenever a pedestrian stands near a curb causes severe congestion and rear-end collision hazards. Utility perception—prominently active in Scenarios 4, 7, 13, 17, and 26—evaluates the degree to which an AV ought to maintain dynamic flow, optimize traffic throughput, and avoid disruptive, hyper-conservative maneuvers while managing ambiguity.

4. Legal Perception

Legal Perception assesses the strictness with which an automated vehicle is expected to comply with codified traffic statutes, jurisdictional regulations, and formal rules of the road. Autonomous systems are designed with rigid programmatic rule sets; however, naturalistic human driving routinely requires pragmatic technical violations of traffic laws (e.g., nudging across a continuous white or yellow line to bypass a double-parked delivery truck). Legal perception—critical in Scenarios 5, 11, 12, and 26—quantifies whether respondents demand absolute fidelity to statutory law or condone flexible rule relaxation to resolve ambiguity safely.

5. Safety Perception

Safety Perception constitutes the fundamental psychological appraisal of physical hazard, collision probability, injury severity, and threat avoidance. As the primary objective of automotive engineering, safety perception operates as an omnipresent baseline across ambiguous scenarios. It captures the user’s perception of risk mitigation, defensive trajectory adjustments, and margins of error. In the factor-analytic model of Baby et al. (2024), Safety Perception was pervasive across nearly all evaluated ambiguous scenarios (functioning across 20 of the 28 scenarios, excluding only those isolated scenarios where ethical or utility dimensions dominated the variance), demonstrating that safety serves as the non-negotiable anchor against which all other perceptual dimensions are balanced.

Theoretical Framework

The Ambiguous Driving Scenarios Survey is situated at the intersection of cognitive ergonomics, normative moral philosophy, and traffic social psychology. Its theoretical foundation draws primarily upon three conceptual paradigms:

1. Normative Ethics and Moral Psychology

The operationalization of the survey draws heavily from classical normative ethical theories, specifically contrasting Deontological Ethics (Kantianism) with Consequentialism / Utilitarianism (Bentham, Mill). In transportation automation, the Moral Machine experiment (Awad et al., 2018) highlighted global cross-cultural variations in how humans resolve unavoidable fatal collisions. However, Baby et al. (2024) recognized that extreme, catastrophic dilemmas represent a vanishingly small fraction of real-world driving. Thus, the authors integrated Dual-Process Theory of moral judgment (Greene et al., 2001), which posits that human evaluations alternate between fast, emotionally driven deontological intuitions (captured by Moral and Legal Perception) and deliberate, cognitive cost-benefit calculations (captured by Utility and Safety Perception). Ambiguous driving scenarios force an interaction between these dual processes because ambiguity creates uncertainty regarding whether a physical hazard will materialize.

2. Theory of Bounded Rationality and Naturalistic Decision Making

Classical engineering models assume ideal driving agents operating under perfect information. In contrast, the survey is grounded in Herbert Simon’s theory of Bounded Rationality and Gary Klein’s Naturalistic Decision Making (NDM). Road users operate in environments with incomplete perceptual data, visual occlusions, fluctuating weather, and unpredictable human intent. When an AV encounters an ambiguous scenario, it cannot calculate deterministic risk probabilities; it must practice ‘satisficing’—selecting an action that meets acceptable thresholds of safety, legality, and social acceptability without freezing traffic. The survey quantifies human expectations of how an artificial cognitive agent should resolve this bounded, satisficing dilemma.

3. Theory of Planned Behavior and Sociotechnical Trust

From an applied perspective, the instrument leverages the Theory of Planned Behavior (Ajzen, 1991) and the Lee and See (2004) framework of Trust in Automation. Human acceptance of autonomous vehicles is not determined solely by the statistical crash rate per million miles, but by the alignment between automated behavior and societal normative expectations. If an AV’s resolution of an ambiguous scenario violates human legal or moral standards (even if mathematically collision-free), human drivers and surrounding road users experience cognitive dissonance, leading to loss of trust, disengagement, or road rage. The survey establishes the psychometric coordinates necessary to engineer AV behaviors that foster appropriate, calibrated trust.

Validity

The Ambiguous Driving Scenarios Survey underwent comprehensive psychometric validation procedures to confirm that its 28 scenarios and five underlying perceptual subscales accurately and consistently measure the target constructs without systematic bias.

Content and Face Validity

Content validity was established through an extensive, multi-tier qualitative and exploratory development pipeline:

• Scenario Generation: The research team conducted an exhaustive synthesis of literature, driving simulator datasets, and crash databases, initially compiling an inventory of over 10,000 candidate driving scenarios.
• Expert Rationalization: Through structured analysis conducted in consultation with automotive engineers, traffic safety specialists, and human-computer interaction (HCI) researchers, this pool was filtered down to 6,590 scenarios by removing redundancies and non-ambiguous situations.
• UX Card Sorting Protocol: To extract the most representative and perceptually salient archetypes of ambiguity, a formal user experience (UX) card sorting study was conducted with active drivers. This qualitative taxonomy reduction condensed the massive scenario database into 28 definitive, pictorial ambiguous scenarios representing critical edge cases in urban, suburban, and rural transit.

Construct and Factorial Validity

Construct validity was evaluated by administering the 28-scenario survey to a representative sample of licensed drivers aged 20 to 49 in the Republic of Korea. The factorial validity was examined through a two-stage structural equation modeling framework combining exploratory and confirmatory techniques:

• An initial Exploratory Factor Analysis (EFA) was used to explore the latent structure of the responses across scenarios.
• A subsequent Confirmatory Factor Analysis (CFA) tested and confirmed the five-factor structural model (Moral, Ethical, Utility, Legal, and Safety Perceptions). Items exhibiting psychometric deficiencies—specifically low standardized factor loadings (< 0.30) or extreme modification indices causing structural strain—were systematically removed (five items dropped during model refinement), yielding an optimized, parsimonious model with excellent overall fit indices.

Convergent and Discriminant Validity

To satisfy the empirical criteria for convergent validity established in structural equation modeling (Fornell & Larcker, 1981; Krishnan & Ramasamy, 2011):

• Convergent Validity: All retained standardized factor loadings for the respective latent constructs significantly exceeded the conventional threshold of 0.60. Furthermore, the Average Variance Extracted (AVE) for each of the five latent constructs surpassed the recommended benchmark of 0.50, demonstrating that the constructs explain more than half of the variance in their designated indicator variables.
• Discriminant Validity: Discriminant validity was verified via the Fornell-Larcker criterion. The square root of the AVE for each construct was strictly greater than the highest bivariate inter-construct correlation with any other latent variable in the model. This statistical separation confirms that Moral, Ethical, Utility, Legal, and Safety perceptions represent conceptually autonomous psychometric dimensions rather than reflections of a single undifferentiated halo effect.

Reliability

The reliability of the Ambiguous Driving Scenarios Survey was rigorously evaluated using both internal consistency metrics and split-half testing procedures across the five latent perceptual dimensions:

Composite Reliability (CR)

Recognizing that Cronbach’s alpha can underestimate reliability in multi-dimensional scales with unequal factor loadings (tau-equivalence violation), the authors calculated Composite Reliability (CR) values for each latent construct. The CR scores across all five dimensions ranged from 0.800 to 0.851, comfortably exceeding the widely accepted psychometric threshold of 0.70 for basic research and 0.80 for advanced applied assessment:

• Moral Perception Construct: Demonstrated high composite reliability (CR within the 0.80–0.85 band), showing high internal coherence in how respondents evaluate care and blameworthiness across ambiguous contexts.
• Ethical Perception Construct: Demonstrated strong internal consistency, confirming consistent normative reasoning regarding fairness and risk distribution.
• Utility Perception Construct: Achieved robust composite reliability, reflecting stable evaluative patterns regarding vehicle efficiency and collective transit flow.
• Legal Perception Construct: Showed high internal coherence regarding statutory adherence across varied driving environments.
• Safety Perception Construct: Maintained strong composite reliability (CR > 0.81), indicating robust measurement stability for collision avoidance and threat appraisal.

Split-Half Reliability

To evaluate the stability of the instrument across alternative halves of the test items and eliminate item-order biases, split-half reliability coefficients were computed. For all five perceptual constructs, the split-half reliability coefficients consistently exceeded 0.812. This high degree of split-half stability confirms that the questionnaire items possess high internal equivalence and that user evaluations are not artifacts of measurement fatigue or item sequence across the 28 visual scenario presentations.

Factor Analysis

The dimensional architecture of the Ambiguous Driving Scenarios Survey was elucidated using a rigorous two-step factor analytic framework comprising initial exploratory modeling followed by confirmatory structural verification.

Exploratory Factor Analysis (EFA)

During the exploratory phase, an EFA utilizing maximum likelihood estimation with oblique rotation was conducted on the participant response matrix across the 28 driving scenarios. The exploratory results indicated that a single-factor ‘general driving appraisal’ model was wholly inadequate, failing to account for the variance across distinct conflict types. Instead, the scree plot and eigenvalue analysis (> 1.0 criterion) clearly delineated a five-dimensional latent space corresponding to Moral, Ethical, Legal, Utility, and Safety dimensions.

Confirmatory Factor Analysis (CFA) and Model Refinement

Confirmatory Factor Analysis was subsequently deployed to test whether the empirically derived five-factor model fit the observed covariance structure. During initial CFA iterations, the research team examined standardized residuals and modification indices (MIs). Five specific items demonstrated psychometric deficiencies—characterized by unacceptable factor loadings (< 0.30) or severe cross-loadings that inflated modification indices. These five items were systematically excised to enhance model parsimony and theoretical clarity.

The refined structural model demonstrated excellent goodness-of-fit indices across standard evaluation criteria (e.g., Comparative Fit Index [CFI] > 0.90, Tucker-Lewis Index [TLI] > 0.90, and Root Mean Square Error of Approximation [RMSEA] < 0.08), confirming that the five-factor structural architecture reliably captures human evaluative variance.

Latent Scenario Mapping

A critical contribution of the CFA in Baby et al. (2024) was the empirical classification of the 28 ambiguous driving scenarios into primary factor loadings. Rather than every scenario loading uniformly across all constructs, the CFA demonstrated that specific types of ambiguity activate distinct normative priorities:

• Moral Scenarios: Scenarios 5, 11, 20, and 22 loaded primarily on the Moral Perception factor. These scenarios typically involve acute human vulnerability, direct eye contact or lack thereof, and potential physical trauma to pedestrians where personal virtue and compassion are paramount.
• Ethical Scenarios: Scenarios 1, 4, 17, and 23 loaded heavily on Ethical Perception. These dilemmas involve competing social rights—such as navigating around vulnerable micromobility users, lane-merging disputes, and situations requiring equitable distribution of discomfort or risk.
• Legal Scenarios: Scenarios 5, 11, 12, and 26 loaded on Legal Perception. These scenarios present acute tensions between strict statutory compliance (e.g., solid lines, stop lines, right-of-way signage) and practical driving progress. Note that Scenarios 5 and 11 cross-load on both Moral and Legal dimensions, reflecting the natural friction between human moral duty and codified traffic law.
• Utility Scenarios: Scenarios 4, 7, 13, 17, and 26 loaded significantly on Utility Perception. These operational contexts involve potential traffic bottlenecks, slow-moving obstacles, and efficiency trade-offs where vehicle deceleration or unnecessary stopping adversely impacts surrounding traffic flow.
• Safety Scenarios: The Safety Perception construct demonstrated an overarching structural role. The CFA revealed that all scenarios loaded significantly on Safety Perception, with the exception of Scenarios 1, 4, 5, 7, 13, 20, 22, and 23 (in which moral, ethical, or utility dilemmas superseded primary physical threat appraisals). Thus, for 20 of the 28 scenarios, physical threat mitigation is the dominant factor driving user perception.

Instrument / Measurement Tool

The Ambiguous Driving Scenarios Survey is structured as an electronic, computer-assisted self-administered questionnaire (CAWI). Below is the comprehensive structural specification of the tool:

• Instrument Type: Psychometric Inventory / Evaluative Scenario-Based Survey.
• Administration Format: Digital / Electronic web-based questionnaire (accessible via PC or tablet display to ensure optimal visual resolution of scenario graphics).
• Target Respondent Group: Adult licensed drivers (validated on ages 20–49, applicable across adulthood 18+).
• Structural Organization: The survey is divided into three sequential sections:
  • Section 1: AV Background & General Perceptions: Introductory briefing defining autonomous vehicles (SAE Level 3+), outlining the purpose of the study, and introducing baseline conceptual definitions of the five evaluative criteria (Moral, Ethical, Legal, Utility, Safety).
  • Section 2: Demographic Profile: Collection of socio-demographic indicators including age, gender, driving experience (years licensed), annual driving mileage, prior familiarity with Advanced Driver Assistance Systems (ADAS), and technological affinity.
  • Section 3: Ambiguous Scenario Evaluation: The core psychometric battery containing 28 distinct ambiguous driving scenario modules presented in randomized or structured order. Each scenario module contains:
    • A standardized pictorial diagram illustrating the spatial geometry, road layout, AV position, surrounding road users, and environmental conditions.
    • A concise textual vignette describing the ambiguous behavioral cue (e.g., a pedestrian standing near a zebra crossing looking at a smartphone without stepping onto the asphalt).
    • Five standardized evaluative rating prompts mapping directly to the five latent factors: Moral, Ethical, Legal, Utility, and Safety importance.
• Total Item Count: 28 scenarios × 5 core perception questions = 140 core scenario-evaluation items, plus baseline demographic and introductory perception items.
• Response Format: Five-point Likert scale anchored as follows:
  • 1 = Very important
  • 2 = Important
  • 3 = Neutral / Moderately important
  • 4 = Slightly important
  • 5 = Not at all important

  (Note: In quantitative statistical analysis, items are typically reverse-coded [1 = Not at all important to 5 = Very important] so that higher numerical values reflect greater perceived importance of that dimension in guiding AV action).

• Scoring and Computational Procedures:
  • Subscale Mean Scores: Calculated by averaging the scores of items corresponding to each latent construct across all 28 scenarios or within scenario clusters.
  • Scenario Priority Profiling: For each individual scenario, a radar-plot profile is generated by plotting the mean importance ratings across Moral, Ethical, Legal, Utility, and Safety dimensions, revealing the scenario’s dominant normative demand.
  • Missing Data Handling: Full completion of electronic prompts is enforced via forced-choice form validation; missing values in field deployments are addressed via full information maximum likelihood (FIML) estimation.

Permissions & Fee and Test Year

The Ambiguous Driving Scenarios Survey was formally published in 2024 in the journal Accident Analysis & Prevention (Baby et al., 2024). The conceptual methodology, statistical findings, and scenario classifications are documented within the scientific publication. However, the complete visual stimulus battery (the 28 standardized pictorial scenario cards and proprietary descriptive item inventories) is not released under an open-source public creative commons license.

Academic researchers, automotive developers, and institutions seeking to deploy the complete, official 28-scenario survey battery or integrate its graphical vignettes into driving simulators must secure formal permission from the corresponding author and copyright holder:

• Corresponding Author: Prof. Seul Chan Lee
• Institution: Division of Media, Culture, and Design Technology, Hanyang University ERICA
• Postal Address: Lion’s Hall 202, 55 Hanyangdaehak-ro, Sangnok-gu, Ansan-si, Gyeonggi-do, Republic of Korea
• Inquiry Email: [email protected]
• Commercial Licensing & Fees: Commercial deployments, industrial algorithm benchmarking, or incorporation into proprietary automotive testing suites may be subject to institutional licensing agreements and copyright permissions administered through Elsevier and Hanyang University ERICA. Academic non-commercial research use generally requires formal written attribution and direct authorization from the primary authors.

References

• Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50(2), 179–211. https://doi.org/10.1016/0749-5978(91)90020-T
• Awad, E., Dsouza, S., Kim, R., Schulz, J., Henrich, J., Shariff, A., Bonnefon, J. F., & Rahwan, I. (2018). The Moral Machine experiment. Nature, 563(7729), 59–64. https://doi.org/10.1038/s41586-018-0637-6
• Baby, T., Ippoliti, H. Ş., Wintersberger, P., Zhang, Y., Yoon, S. H., Lee, J., & Lee, S. C. (2024). Development and classification of autonomous vehicle’s ambiguous driving scenario. Accident Analysis & Prevention, 200, Article 107501. https://doi.org/10.1016/j.aap.2024.107501
• Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39–50. https://doi.org/10.1177/002224378101800104
• Greene, J. D., Sommerville, R. B., Nystrom, L. E., Darley, J. M., & Cohen, J. D. (2001). An fMRI investigation of emotional engagement in moral judgment. Science, 293(5537), 2105–2108. https://doi.org/10.1126/science.1062872
• Krishnan, J., & Ramasamy, M. (2011). A study on the factors influencing consumer buying behavior of luxury cars. Journal of Contemporary Research in Management, 6(1), 89–102.
• Lee, J. D., & See, K. A. (2004). Trust in automation: Designing for appropriate reliance. Human Factors, 46(1), 50–80. https://doi.org/10.1518/hfes.46.1.50.30392

Items of the Scale