Occlusion-induced risk and interventions in pedestrian-autonomous truck interactions on multi-lane roads: A virtual reality study

1. Introduction

• Truck-specific physical attributes and automation-related uncertainty create direct interaction risks, while the truck body can conceal adjacent traffic and create secondary occlusion risks.The introduction identifies truck size, braking capability, communication cues, and visibility of far-side vehicles or conflict areas as relevant risk mechanisms.Source: §1 · p. 3
• Conventional eHMIs mainly communicate the ego vehicle’s awareness, motion, or yielding intention, so the introduction calls for interfaces that reveal hidden hazards and localize risk; auditory warnings may add a non-visual channel.The cited prior-work claims are attributed to De Clercq et al. (2019), Tabone et al. (2023), Ye et al. (2026b), Soares et al. (2021), and Bindschädel et al. (2023).Source: §1 · pp. 3–4
• In multi-lane crossings, risk may shift from near-side truck interaction to far-side vehicles; rain-related braking reductions and far-side yielding strategy may also affect safety margins and crossing decisions.The study therefore examines near-side vehicle type, weather, and far-side AV interaction strategy.Source: §1 · p. 4
• The study proposes and tests an occlusion-aware mitigation strategy that extends beyond vehicle-intention communication to support occlusion awareness, hidden-risk recognition, and spatial risk localization.The introduction frames this as the study’s main contribution to pedestrian–autonomous truck interaction safety.Source: §1 · p. 4
• A controlled VR experiment examines how near-side vehicle type, weather, and far-side AV interaction strategy affect pedestrian behavior, perceived risk, and objective safety.The study aims to identify both direct pedestrian–truck risk and occlusion-induced secondary risk.Source: §1 · p. 4
• Three interventions—an environment-aware eHMI, projected eHMI, and auditory warning—are tested in a representative high-risk scenario using perceived risk, behavior, DST, and ITADV as evaluation measures.The evaluation asks whether occlusion-related information supports behavioral adjustment and improves objective safety margins.Source: §1 · pp. 4–5

2.1 Risk formation in pedestrian-ATs interaction

• ATs create two connected pedestrian risks: direct interaction shaped by braking limits and missing driver cues, and secondary hidden-traffic risk from occlusion, whose combined effects remain underexplored.Large truck dimensions and constrained stopping capability contribute to direct risk, while truck bodies can conceal approaching vehicles and conflict sources.Source: §2.1 · pp. 5–6

2.2 Behavioral and risk indicators in pedestrian-AVs interaction

• Behavioral measures describe crossing decisions and movement, while TTC, PET, DST, and ITADV capture complementary conflict, braking-capability, and occlusion-related time dimensions needed for pedestrian-AT safety.The review argues that behavioral indicators and surrogate safety measures provide complementary perspectives.Source: §2.2 · p. 6

2.3 Interventions for pedestrian-AVs interaction safety

• Conventional vehicle-mounted eHMIs can communicate status or yielding and affect crossing decisions, but their effectiveness depends on design, familiarity, vehicle motion, and traffic context.In AT scenarios, pedestrians may need information about both the truck’s state and occluded traffic.Source: §2.3 · p. 7
• Projected eHMIs can place warnings near conflicts, whereas auditory cues attract attention but may poorly localize risk; whether either improves occlusion awareness and objective safety in AT settings remains unclear.AT interventions may need to communicate truck intent while also supporting hidden-risk recognition and spatial risk localization.Source: §2.3 · pp. 7–8

3. Methods

• The two-stage controlled VR study examined risk formation without interventions and validated targeted measures in a predefined high-risk scenario, with all trials randomized together.The high-risk case combined near-side truck occlusion, rainy-weather braking constraints, and a non-yielding far-side vehicle.Source: §3 · p. 8

3.1 Experiments: platform and apparatus

• The CAVE-based platform combined head-mounted VR, interactive movement control, dynamic vehicle simulation, and synchronous recording to support repeatable pedestrian–vehicle interactions.The system recorded pedestrian movement, vehicle trajectories, and their spatiotemporal relationships.Source: §3.1 · p. 9

3.2.1 Experiments scenarios

• The unsignalized two-lane crossing placed two near-side vehicles as both direct interaction and occluding vehicles, followed by a far-side passenger AV whose yielding strategy varied.Vehicles began at 50 km/h; the near-side vehicles stopped before the crossing path, while the far-side AV yielded or maintained priority.Source: §3.2.1 · p. 10

3.2.2 Parametric design; deceleration parameters

• The risk-formation design crossed near-side vehicle type, sunny or rainy weather, and far-side yielding or non-yielding strategy, with braking capability varied by vehicle and weather.Equivalent braking distances were 11.21 and 12.36 m for passenger AVs in sunny and rainy conditions, and 18.01 and 22.05 m for ATs, respectively.Source: §3.2.2 · pp. 11–12, deceleration parameters · p. 12

3.2.3 Interventions

• Three measures targeted truck-induced occlusion and far-side risk localization: a truck-mounted environment-aware eHMI, a road-surface projected eHMI, and a headset auditory warning.The interventions were designed for the AT–rainy–non-yielding scenario, with the audio alert providing a non-visual channel when visual information might be missed.Source: §3.2.3 · p. 13

3.3 Experiment procedure

• Each of 54 participants completed 11 randomized within-subject trials—eight risk-formation conditions and three intervention trials—and all completed the study without reported sickness symptoms.Participants provided an immediate perceived-risk rating after each crossing, and the system recorded crossing behavior and pedestrian–vehicle relations.Source: §3.3 · p. 15

3.4.1 Demographic Variables

• The sample comprised 54 campus recruits, including 22 women and 32 men aged 20–29 years; demographic characteristics included education and driving experience.Mean age was 24.04 years (SD 1.58); 41 participants had a bachelor’s degree or lower and 13 had a master’s degree or higher.Source: §3.4.1 · pp. 16–17

3.4.2 Independent variables; 3.4.3 Dependent variable

• Risk-formation predictors were the three binary factors in a 2 × 2 × 2 repeated-measures design; intervention analyses compared three measures with no intervention as the reference.Perceived risk was rated on a 10-point scale separately for first- and second-lane interactions; intervention analyses also included behavioral indicators, alongside objective risk measures.Source: §3.4.2 · p. 17, §3.4.3 · p. 18

3.4.3 Dependent variable

• Objective risk was measured separately for each lane: DST for the first-lane conflict and ITADV for the second-lane conflict.Source: §3.4.3 · p. 19
• DST is the ratio of required collision-avoidance deceleration to actual average deceleration; negative values indicate lower first-lane risk, while values near or above zero indicate higher risk.DST is sensitive to vehicle dynamics and weather-related changes in braking efficiency.Source: §3.4.3 · p. 19
• ITADV, expressed in seconds, captures pedestrian time advantage over vehicles with a minimum safety-time threshold; larger values mean greater separation, and it targets far-side risk under near-side occlusion.Source: §3.4.3 · p. 20
• Intervention responses were tracked through waiting, crossing, and vehicle-observation times, representing hesitation and timing, exposure and movement, and gaze-based attention, respectively.Vehicle-observation time was derived from VR headset eye-tracking frames in which gaze fell within the relevant vehicle-body area; the indicators helped distinguish increased risk awareness, changes to crossing, and objective safety-margin improvements.Source: §3.4.3 · pp. 20–21

3.5 Statistical modeling

• The analysis combined descriptive summaries, mixed models, and mediation analysis to examine pedestrian–AT risk and intervention effects.Mediation analysis tested whether second-lane entry timing transmitted the AT effect on second-lane objective risk or whether an occlusion-related effect remained.Source: §3.5 · p. 21

3.5.1 Descriptive statistics

• For each lane and condition, descriptive statistics reported means, standard deviations, minima, and maxima for perceived risk, behavioral indicators, and objective risk measures.Comparisons across experimental conditions and lanes provided an initial overview before model-based analysis.Source: §3.5.1 · p. 21

3.5.2 Linear mixed models

• Participant-random-intercept LMMs accounted for repeated observations; risk models tested vehicle type, weather, far-side strategy, and vehicle-type-by-weather interaction, while intervention models used no intervention as the reference.The intervention models evaluated perceived risk, behavioral indicators, and objective safety measures.Source: §3.5.2 · pp. 21–22

3.5.3 Mediation analysis

• Mixed-effects mediation decomposed the near-side AT effect on second-lane ITADV into an entry-time indirect pathway (ab) and a remaining direct occlusion effect, adjusting for weather, far-side strategy, gender, driving experience, and education.Lower ITADV indicates a smaller temporal advantage and higher objective risk.Source: §3.5.3 · pp. 22–24
• Indirect effects were evaluated with 5,000 participant-level cluster-bootstrap replications and bias-corrected 95% confidence intervals, with significance defined by intervals excluding zero.Significant indirect and direct effects indicated partial mediation.Source: §3.5.3 · pp. 24–25

4.1.1 Descriptive results of the risk formation experiment; 4.1.2 Descriptive results of the intervention-effect experiment

• Descriptives showed phase-dependent risk in the formation experiment, while intervention trials selected a critical context with near-boundary DST, low ITADV, and high perceived risk.Second-lane perceived risk exceeded first-lane risk in the formation experiment; intervention-trial perceived risk was high in both lanes.Source: §4.1.1 · p. 25, §4.1.2 · p. 26

4.2.1 Risk perception in the first lane

• ATs and rainy weather each increased first-lane perceived risk, indicating that pedestrians responded to truck-related risk cues and adverse conditions during direct interaction.The AT effect was β = 3.046 (p < 0.001); the weather effect was also significant (p < 0.001).Source: §4.2.1 · p. 27

4.2.2 Objective risk in the first lane

• ATs reduced first-lane DST by 0.290, but the vehicle-type-by-weather interaction of 0.383 showed that rain weakened this larger temporal safety margin.The pattern suggests pedestrian risk compensation under normal conditions, becoming less effective when rain reduces braking efficiency.Source: §4.2.2 · p. 28

4.2.3 Risk perception in the second lane

• In the second lane, perceived risk rose with ATs (β = 1.009), rain (β = 0.370), and far-side vehicle interaction strategy (β = 0.778).The random-intercept model offered limited evidence of stable between-participant differences in baseline second-lane perceived risk.Source: §4.2.3 · p. 29

4.2.4 Objective risk in the second lane

• ATs reduced second-lane ITADV by 1.494; entry timing mediated 27.10% of this effect, while the remaining direct occlusion effect accounted for 72.90%.The larger truck body continued to restrict recognition of the far-side vehicle after accounting for entry timing.Source: §4.2.4 · pp. 30–31

4.3.1 Effects of intervention measures on risk perception

• All three interventions increased perceived risk in both lanes; projected eHMI led, with increases of 0.889 in lane 1 and 0.870 in lane 2.Auditory warning and environment-aware eHMI also increased perceived risk, but less than projected eHMI in each lane.Source: §4.3.1 · p. 32

4.3.2 Effects of intervention measures on objective risk

• Projected eHMI alone improved both objective safety margins: DST fell by 0.142 in lane 1 and ITADV rose by 0.255 in lane 2.Neither auditory warning nor environment-aware eHMI significantly affected either objective indicator.Source: Table 13 · p. 32, §4.3.2 · pp. 32–33

4.3.3 Effects of intervention measures on pedestrian behavior

• Environment-aware eHMI increased first-lane waiting and observation times while reducing crossing time, prompting longer observation before crossing and quicker movement once started.The changes were +1.204, +0.613, and −0.388, respectively.Source: §4.3.3 · p. 33
• Behavioral changes were not accompanied by significant improvements in DST or ITADV.The passage describes the changes as part of a cautious pre-crossing process.Source: §4.3.3 · p. 33
• −0.952 (β, p < 0.01) was the projected eHMI effect on first-lane vehicle observation time, while 0.222 (β, p < 0.01) was its effect on second-lane crossing time.The shorter observation time suggests spatially localized information helped pedestrians identify the relevant risk source faster; the longer second-lane crossing time indicated a slower, potentially more cautious traversal.Source: §4.3.3 · p. 33
• Auditory warning did not significantly affect waiting time, crossing time, or vehicle observation time.Its influence was concentrated on perceived risk rather than observable crossing behavior.Source: §4.3.3 · p. 33

5.1 How ATs shape pedestrian interaction risk

• ATs and rainy weather each increased perceived risk, while the larger temporal safety margin associated with ATs reflected cautious crossing rather than inherent truck safety.The authors interpret this pattern as risk compensation: pedestrians perceived ATs as more threatening and adopted more conservative crossing timing.Source: §5.1 · p. 34
• Rain weakened the AT-associated temporal safety margin by constraining braking, although pedestrians did not fully internalize how the AT–rain combination changed objective safety.The discussion attributes this mismatch to reduced braking efficiency and the less directly perceptible dynamic consequences of wet conditions.Source: §5.1 · pp. 34–35

5.2 How ATs affect pedestrian interaction with other vehicles through occlusion

• ATs significantly reduced second-lane ITADV; mediation results indicate that compressed crossing timing and a remaining direct effect both contributed to increased far-side interaction risk.The discussion links the remaining effect to occlusion that limits recognition of the far-side vehicle’s position, speed, and arrival timing.Source: §5.2 · pp. 35–36

5.3 Intervention effectiveness

• Projected eHMI significantly reduced DST and increased ITADV, with favorable behavioral adjustment, apparently by locating warnings at the potential conflict area.The discussion suggests that this spatial connection helps pedestrians localize risk under occlusion.Source: §5.3 · pp. 36–37
• Environment-aware eHMI increased perceived risk and changed waiting and observation behavior without improving objective safety, while auditory warnings raised perceived risk but had unstable objective effects.The discussion suggests that vehicle-mounted messages may focus attention on the truck, while auditory warnings have weak spatial directivity and may be better suited to supplementary risk arousal.Source: §5.3 · p. 37

5.4 Practical implications

• For rain, low-adhesion pavement, or reduced visibility, the discussion recommends more conservative AT operation, including lower approach speeds, earlier deceleration, and larger yielding and stopping margins.It also proposes high-caution modes for unsignalized multi-lane crossings and treating such scenarios as priority monitoring areas.Source: §5.4 · p. 38
• Safety management and vehicle evaluation should consider whether AT occlusion impairs recognition of subsequent risk sources and changes pedestrians’ decisions and safety margins, not only whether the truck yields.The discussion frames occlusion as a broader interaction risk beyond a single vehicle-pedestrian encounter.Source: §5.4 · pp. 38–39
• AT interfaces should help pedestrians locate potential conflict areas, with spatially directed warnings connecting information to the hazard rather than focusing attention only on the vehicle surface.The discussion highlights projected eHMI’s transfer of risk information from the vehicle surface to the conflict space as potentially well suited to occlusion scenarios.Source: §5.4 · p. 39

6. Conclusions

• ATs raised perceived risk but encouraged cautious crossing that increased temporal safety margins under some conditions; rain weakened this compensation, and truck occlusion contributed to reduced second-lane ITADV.The mediation analysis linked the second-lane effect to both entry timing and occlusion from the truck’s large body.Source: §6 · pp. 39–40
• Projected eHMI performed best overall, improving DST, significantly increasing ITADV, and supporting behavior aligned with risk localization; the other interventions improved risk awareness but showed weaker safety benefits.Environment-aware eHMI did not produce consistently favorable objective-safety outcomes, while auditory warning showed limited evidence of behavioral or objective-safety improvement.Source: §6 · p. 40
• The study’s limitations include VR’s imperfect realism, a mainly young university-student sample, and focus on one unsignalized two-lane crossing scenario.Future research should test more road environments, participant groups, vehicle configurations, speeds, and combined interventions to validate occlusion-targeted safety support.Source: §6 · p. 40

Authorship contribution statement

• Yun Ye, Yuan Che, S.C. Wong, Stergios-Aristoteles Mitoulis, and Haoyang Liang contributed to the study and manuscript in the roles listed in the authorship statement.Roles included conceptualization, data curation, analysis, investigation, methodology, supervision, funding, and writing.Source: Authorship contribution statement · p. 40

Declaration of the use of AI

• The authors used ChatGPT to improve manuscript readability and language, then reviewed and edited the text and accepted full responsibility for the publication.Source: Declaration of the use of AI · p. 40

Ethical statement

• The study protocol received prior approval from the Science and Technology Ethics Committee of Tongji University, approval number tjdxsr2024041.Source: Ethical statement · pp. 40–41

Data availability statement

• The authors state that data are available on reasonable request.Source: Data availability statement · p. 41

Declaration of competing interest

• The authors declared no known competing financial interests or personal relationships that could have influenced the reported work.Source: Declaration of competing interest · p. 41

Appendix A1. Virtual reality sickness questionnaire (VRSQ)

• The questionnaire includes general discomfort and fatigue.Both are listed as symptoms in Appendix A1.Source: Appendix A1. Virtual reality sickness… · p. 49
• Visual symptoms include eyestrain and difficulty focusing.Source: Appendix A1. Virtual reality sickness… · p. 49
• The list also includes headache and fullness of head.Source: Appendix A1. Virtual reality sickness… · p. 49
• Blurred vision is another listed symptom.Source: Appendix A1. Virtual reality sickness… · p. 49
• The questionnaire asks about dizziness with eyes closed.Source: Appendix A1. Virtual reality sickness… · p. 49
• Vertigo is also included in the symptom list.Source: Appendix A1. Virtual reality sickness… · p. 49

• Weather and near-side vehicle type are listed as condition-wise dimensions.Source: Appendix A2. Condition-wise… · p. 49
• Far-side strategy is listed alongside DST, ITADV, and risk perception (Lane 1).Source: Appendix A2. Condition-wise… · p. 49

Appendix A4. Total-effect model (Dependent variable: ITADV, path c)

• The autonomous-truck coefficient for ITADV was −1.494 (p < 0.001; 95% CI [−1.802, −1.186]).The model intercept was 6.465 (p = 0.010; 95% CI [1.514, 11.416]).Source: Appendix A4. Total-effect model… · pp. 50–51
• Rainy weather (−0.046, p = 0.768) and a non-yielding far-side strategy (−0.246, p = 0.118) were not significant predictors of ITADV.Their respective 95% CIs were [−0.354, 0.262] and [−0.554, 0.062].Source: Appendix A4. Total-effect model… · p. 50
• Gender had a coefficient of 0.502 (p = 0.112), while age had a coefficient of −0.191 (p = 0.067); neither was significant at p < 0.05.The 95% CIs were [−0.117, 1.120] for gender and [−0.395, 0.014] for age.Source: Appendix A4. Total-effect model… · p. 50
• Relative to 0 years of driving experience, the reported experience-category coefficients were −0.379, −0.016, −0.166, and 0.406; none was significant.The first three estimates and their p-values appear in the main table passage; the continuation reports >More than 5 years as 0.406 (p = 0.456).Source: Appendix A4. Total-effect model… · pp. 50–51
• In the mediator-model passage, autonomous-truck status had a coefficient of 1.004 (p < 0.001), rainy weather 0.272 (p = 0.042), and gender −1.262 (p = 0.022).The non-yielding far-side strategy coefficient was −0.068 (p = 0.612); the other listed covariates were also not significant at p < 0.05.Source: Appendix A4. Total-effect model… · p. 51
• Master’s degree or higher had a coefficient of −0.010 (p = 0.988), with a bachelor’s degree or lower as the reference category.The model continuation also reports the >More than 5 years estimate as 0.406 (p = 0.456).Source: Appendix A4. Total-effect model… · p. 51

Appendix A6. Direct-effect model (Dependent variable: ITADV, paths b and c’)

• Autonomous truck (coefficient −1.089, p<0.001) and second-lane entry time (−0.403, p<0.001) had significant negative coefficients in the ITADV model.Both 95% confidence intervals excluded zero: [−1.387, −0.792] for autonomous truck and [−0.484, −0.323] for second-lane entry time.Source: Appendix A6. Direct-effect model… · p. 52
• Rainy weather (0.064, p=0.665) and a non-yielding far-side strategy (−0.273, p=0.062) were not statistically significant predictors.The far-side strategy estimate was near, but above, the conventional 0.05 significance threshold.Source: Appendix A6. Direct-effect model… · p. 52
• Gender, age, and driving-experience categories also had no statistically significant coefficients in the reported model.Reported p-values were 0.976 for gender, 0.254 for age, and 0.111, 0.981, 0.377, and 0.451 across the listed experience categories.Source: Appendix A6. Direct-effect model… · p. 52

Appendix A7. Random effects and model fit

• The direct-effect model had a random-intercept variance of 0.183, log likelihood of −806.828, Wald χ² of 220.85 (p<0.001), and LR test of 6.41 (p<0.01).The reported intercept was 6.604 (SE 1.931, p=0.001; 95% CI [2.820, 10.389]).Source: Appendix A6. Direct-effect model… · pp. 52–53