INTRODUCTION:The electrification of personal transport has transformed urban mobility, but the rapid adoption of e-bikes and e-scooters has introduced distinct fatal crash risks. Existing research on micromobility safety is often limited to non-fatal injuries and relies on standard crash databases that often lack the granular detail needed, for instance, to distinguish between vehicle types, ownership (private vs. rental), or to quantify the severity of alcohol intoxication. METHOD:We conducted a retrospective analysis of all fatal crashes involving conventional bicyclists (n = 152), e-cyclists (n = 34), and e-scooterists (n = 18) recorded in Sweden's unique in-depth fatal crash database (2016-2024). This national-level data, compiled by multidisciplinary teams, allowed for an unprecedented comparative analysis of crash typologies, vehicle characteristics, and rider profiles. RESULTS:The three micromobility modes showed different fatal crash profiles. Conventional bicyclists were old (median age 71.0) involved in multi-road-user crashes during weekdays. In contrast, e-scooterist fatalities involved middle-aged riders (median age 47.5) in single-rider crashes, occurred on weekends and at night, and showed a high prevalence of alcohol intoxication (44.4%). Interestingly, the majority of e-scooterist crashes (66.7%), particularly those involving alcohol, occurred on privately-owned vehicles. E-cyclists occupied an intermediate crash and rider profile, sharing characteristics with both modes. Across all modes, head injuries were the dominant cause of death, while helmet use was critically low or absent. CONCLUSIONS:The unique crash profiles suggest that a vehicle-agnostic regulatory approach may be a missed opportunity to develop appropriate safety interventions. The findings highlight that safety interventions must extend beyond shared fleets to ensure private e-scooterists are not overlooked. The high prevalence of severe alcohol intoxication and lack of helmet use indicate clear areas for intervention. PRACTICAL APPLICATIONS:This study provides a detailed, evidence-based resource for policymakers to develop targeted regulations, safer infrastructure, and create awareness campaigns that address the risks unique to different micromobility modes.

This paper introduces a novel micromobility simulator that combines a large treadmill with stereophotogrammetry to enable virtual testing of real bicycles and e-scooters while preserving the balancing task and allowing lateral maneuvering. Unlike existing riding simulators, the platform accommodates real vehicles, increasing ecological validity for research on balance and control. The simulator architecture is presented and its potential demonstrated by comparing lateral control during obstacle avoidance—a critical task that previous simulators could not address with comparable realism.Because full balance control is preserved, we propose new metrics for comparing balance across micromobility vehicles and show how they complement traditional indicators of lateral control and performance. Overall, the paper contributes 1) a vehicle-agnostic riding simulator and 2) a minimal yet discriminative set of indicators for maneuvering and balance across tasks and vehicles.Twelve participants performed cruising and obstacle-avoidance tasks at different speeds on both a bicycle and an e-scooter. Four indicators were analyzed: standard deviation of lane position and steering angle, adapted from driving simulation research, and standard deviation of lean angle and relative upper–lower body angle, inspired by motor control literature. Results revealed distinct patterns across tasks, speeds, and vehicles, with low redundancy among indicators. Notably, during obstacle avoidance, participants exhibited different postural strategies on bicycles and e-scooters and collided more frequently when cycling.The simulator enables new research on human–vehicle interaction, ergonomics, and safety in micromobility, although further validation and enhancements are needed to fully exploit its potential.

Cycling has gained increasing popularity across Europe, yet the frequency and severity of cyclist-vehicle conflicts at unsignalized intersections remain key road-safety concerns. This study investigates the interaction between drivers and cyclists in such settings, focusing on the role of intersection visibility (IV), difference in time to arrival (DTA) of the car and bicycle, and drivers' gaze behavior in shaping yielding decisions, braking patterns, and speed profiles. Using a driving simulator equipped with eye-tracking technology, participants completed multiple drives through the digital twin of a real-world intersection. The IV was systematically varied by repositioning a parked truck, while the DTA was controlled by triggering the virtual cyclist's approach at different temporal offsets relative to the car's arrival.Mixed-effects Bayesian regression models revealed that both IV and DTA significantly influenced the drivers' likelihood of yielding: higher visibility and a shorter time difference between vehicle and cyclist arrivals consistently increased yielding rates. Gaze behavior also emerged as a critical factor; earlier fixation on the crossing cyclist strongly correlated with the likelihood of deciding to yield. In contrast, no single predictor significantly explained the distance at which drivers initiated braking. Speed-profile analyses further underscored the finding that drivers' deceleration strategies are shaped by visibility constraints and perceived temporal pressure from oncoming cyclists.These findings highlight the importance of visibility, temporal cues, and visual attention metrics in intersection designs and advanced driver assistance systems. Safety technologies and automated features can more accurately anticipate driver-cyclist interactions when gaze behavior is integrated into their predictive models. Future work should confirm these insights through on-road studies, as well as exploring additional intersection layouts and environmental conditions to obtain more data that can lead to enhance both infrastructure design and automated vehicle algorithms.

Naturalistic data, i.e. data collected in real traffic by road users attending their daily routines, are the gold standard for crash causation analyses. In fact, these data can show the pre-crash road-user behaviour that is hard to observe from other crash data. Naturalistic data from 6868 trips by 4694 distinct participants, collected over a period of 1.5 years from 17 e-scooters, were used to estimate crash risk by means of odds ratios (OR) and crash prevalence by population attributable risk percentage (PARP). We computed OR and PARP, comparing crashes and near-crashes to baseline events from normal riding. The baselines were selected through both matching and random sampling strategies in order to expand and increase the statistical significance of previous results-while also providing new methodological insights for future research on crash causation. This study also investigated the impact of different baseline-to-safety-critical event ratios for the assessment of crash risk. From a safety perspective, our findings suggest that safety interventions that reduce leisure trips, exposure to intersections, trips on Fridays and Saturdays, pack riding, and inexperienced riding should be prioritised. From a methodological perspective, we showed how combining random and matched baselines can help quantify the crash risk and crash prevalence for micromobility vehicles. The results from this study may encourage policymakers to make data-driven decisions regarding e-scooter regulations. Future research should combine data from naturalistic studies and crash databases with data from the perspective of other road users to provide a more holistic view of e-scooter safety.

According to crash data reports, most collisions between cyclists and motorized vehicles occur at unsignalized intersections (where no traffic lights regulate vehicle priority). In the era of automated driving, it is imperative for automated vehicles to ensure the safety of cyclists, especially at these intersections. In other words, to safely interact with cyclists, automated vehicles need models that can describe how cyclists cross and yield at intersections. So far, only a few studies have modeled the interaction between cyclists and motorized vehicles at intersections, and none of them have explored the variations in interaction outcomes based on the type of drivers involved. In this study, we compare non-professional drivers (represented by passenger car drivers) and professional drivers (truck and taxi drivers). We also introduce a novel application of game theory by comparing logit and game theoretic models’ analyses of the interactions between cyclists and motorized vehicles, leveraging naturalistic data. Interaction events were extracted from a trajectory dataset, and cyclists’ non-kinematic cues were extracted from videos and incorporated into the interaction events’ data. The modeling outputs showed that professional drivers are less likely to yield to cyclists than non-professional drivers. Furthermore, the behavioral game theoretic models outperformed the logit models in predicting cyclists’ crossing decisions.

INTRODUCTION:Recently, e-scooters have proliferated worldwide. Municipalities have been struggling with regulating e-scooters due to public concerns that the injuries from the new crashes outweigh the health and environmental benefits of micromobility use. Indeed, several studies have reported crash risk for e-scooters 4 to 10 times higher than that for bicycles. METHOD:We had unprecedented access to crash and exposure data collected in 2022 and 2023 from a rental service of e-scooters and e-bicycles in seven European cities. We conducted a retrospective cross-sectional study to compute injury rates and incidence-rate ratios for each city while directly controlling for geography, ownership, and exposure (measured in three different ways). RESULTS:We analyzed 686 e-scooterist and 35 e-cyclist crashes. Injury rates were higher for e-cyclists than e-scooterists in most of the cities, for all exposure measures. Further, the incidence-rate ratios indicate that the injury risk was 2.5-10 times lower for e-scootering than e-cycling. CONCLUSIONS:E-scootering may not be riskier than cycling as several studies have claimed. In fact, by exploiting technology to control for location, exposure, ownership, and usage, our analysis shows that e-scooterists experience lower crash rates than e-cyclists. While our analysis has some limitations and cannot be considered conclusive evidence, taking location, usage, ownership, and high-resolution exposure into account-which our analysis did contrary to previous studies-is crucial for a more accurate comparison among (micromobility) transport modes. In general, our research suggests incorporating geofencing and GPS-derived exposure metrics in future safety assessments. PRACTICAL APPLICATION:The results and methodologies presented in this paper may help urban planning of rental micromobility services within cities.

Introduction: With cycling gaining more popularity in urban areas, it is vital to obtain accurate knowledge of cyclists’ behavior to develop behavioral models that can predict the cyclist’s intent. Most conflicts between cyclists and vehicles happen at crossings where the road users share the path, especially at unsignalized intersections. However, few studies have investigated and modeled the interaction between cyclists and vehicles at unsignalized intersections. Method: A bike simulator experiment was conducted to scrutinize cyclists’ response process as they interacted with a passenger car at an unsignalized intersection. An existing unsignalized intersection in Gothenburg was simulated for test participants. Two independent variables were varied across trials: the difference in time to arrival at the intersection (DTA) and intersection visibility (IV). Subjective and quantitative data were analyzed to model the cyclists’ behavior. Results: When approaching the intersection, cyclists showed a clear sequence of actions (pedaling, braking, and head turning). The distance from the intersection at which cyclists started braking was significantly affected by the two independent variables. It was also found that DTA, looking duration, and pedaling behavior significantly affected cyclists’ decisions to yield. Finally, the questionnaire outputs show that participants missed eye contact or communication with the motorized vehicle. Conclusions: The kinematic interaction between cyclists and vehicles, along with the cyclist’s response process (visual and kinematic), can be utilized to predict cyclists’ yielding decision at intersections. From the infrastructural perspective, enhancing visibility at intersections has the potential to reduce the severity of interactions between cyclists and vehicles. The analysis of the questionnaire emphasizes the significance of visual communication between cyclists and drivers to support the cyclist’s decision-making process when yielding. Practical applications: The models can be used in threat assessment algorithms so that active safety systems and automated vehicles can react safely to the presence of cyclists in conflict scenarios.

Artificial test environments have become indispensable for modeling human behavior in traffic because they provide more control of experimental factors and fewer ethical concerns than a real traffic environment. These advantages are particularly critical for research on interactions involving vulnerable road users such as cyclists and drivers of motorized vehicles. However, previous research comparing test environments has predominantly focused on car-to-car interaction scenarios. This study investigated a cyclist-overtaking scenario in three different environments: a driving simulator (SIM: N = 25), a test track (TT: N = 18), and a hybrid “driver-vehicle-in-the-loop” environment (DVIL: N = 33) in which the participants drove a real vehicle on a test track while wearing a virtual-reality headset, so they could only see the virtual environment. Overall, the results verified that the direction of the main factors affecting overtaking strategy and performance was similar across all environments. In presence of a close oncoming vehicle, drivers in all environments preferred the more cautious accelerative (waiting behind the cyclist) over flying (passing the cyclist while facing the oncoming vehicle) overtaking strategy, and this effect was strongest in SIM. Metrics related to safety margins to the cyclist followed a similar trend in all environments in the presence of oncoming traffic; however, the TT environment had the smallest overall safety margins, suggesting that virtual visual cues induce safer behavior. Because overtaking maneuvers depend on depth estimation, the visual resolution of the virtual environment may be more important than the environment itself in explaining the results.

In recent years, micromobility has seen unprecedented growth, especially with the introduction of dockless e-scooters. However, the rapid emergence of e-scooters has led to an increase in crashes, resulting in injuries and fatalities, highlighting the need for in-depth analysis to understand the underlying mechanisms. While helpful in quantifying the problem, traditional crash database analysis cannot fully explain the causation mechanisms, e.g., human adaptation failures leading to safety-critical events. Naturalistic data have proven extremely valuable for understanding why crashes happen, but most studies have addressed cars and trucks. This study is the first to systematically analyze factors contributing to crashes and near-crashes involving rental e-scooters in an urban environment, utilizing naturalistic data. The collected dataset included 6868 trips, covering 9930 km over 709 h with 4694 unique participants. We identified 61 safety-critical events, including 19 crashes and 42 near-crashes, and subsequently labeled variables associated with each event according to the codebook using video data. Our odds ratio analysis identified that rider experience and behavior (e.g., phone usage, single-handed riding, and pack riding) significantly increase the crash risk. Given the accessibility of rental e-scooters to individuals regardless of their experience, our findings emphasize the need for rider training in addition to education. Influenced by their experience with bicycles, riders may anticipate a similar self-stabilizing mechanism in e-scooters. We found that single-handed riding, which compromises balance, poses a heightened risk, underscoring the crucial role of balance in safe e-scooter operation. Furthermore, the purpose (leisure or commute) and directness (point-to-point or detour) of the trip were also identified as factors influencing the risk, suggesting that user intent plays a role in safety-critical events. Interestingly, our analysis underscores the importance of adapting the crash and near-crash definitions when working with two-wheeled vehicles, especially those in the shared mobility system.

For cyclists, being overtaken represents a safety risk of possibly being side-swiped or cut in by overtaking drivers. For drivers, such maneuvers are challenging-not only do they need to decide when to initiate the maneuver, but they also need to time their return well to complete the maneuver. In the presence of oncoming traffic, the problem of completing an overtaking maneuver extends to balancing head-on with side-swipe collision risks. Active safety systems such as blind-spot or forward-collision warning systems, or, more recently, automated driving features, may assist drivers in avoiding such collisions and completing the maneuver successfully. However, such systems must interact carefully with the driver and prevent false-positive alerts that reduce the driver's trust in the system. In this study, we developed a driver-behavior model of the drivers' return onset in cyclist-overtaking maneuvers that could improve such a safety system. To provide cumulative evidence about driver behavior, we used data from two different sources: test track and naturalistic driving. We developed Bayesian survival models for the two datasets that can predict the probability of a driver returning, given time-varying inputs about the current situation. We evaluated the models in an in-sample and out-of-sample evaluation. Both models showed that drivers use the displacement of the cyclist to time their return decision, which is accelerated if an oncoming vehicle is present and close. We discuss how the models could be integrated into an active-safety system to improve driver acceptance.

Objective This study aims to understand drivers' visual attention before and after take-over requests during automated driving (AD), when the vehicle is fully responsible for the driving task on public roads. Background Existing research on transitions of control from AD to manual driving has mainly focused on take-over times. Despite its relevance for vehicle safety, drivers' visual attention has received little consideration. Method Thirty participants took part in a Wizard of Oz study on public roads. Drivers' visual attention was analyzed before and after four take-over requests. Visual attention during manual driving was also recorded to serve as a baseline for comparison. Results During AD, the participants showed reduced visual attention to the forward road and increased duration of single off-road glances compared to manual driving. In response to take-over requests, the participants looked away from the forward road toward the instrument cluster. Levels of visual attention towards the forward road did not return to the levels observed during manual driving until after 15 s had passed. Conclusion During AD, drivers may look toward non-driving related task items (e.g., mobile phone) instead of forward. Further, when a transition of control is required, drivers may take over control before they are aware of the driving environment or potential threat(s). Thus, it cannot be assumed that drivers are ready to respond to events shortly after the take-over request. Application It is important to consider the effect of the design of take-over requests on drivers' visual attention alongside take-over times.

Introduction: While micromobility vehicles offer new transport opportunities and may decrease fuel emissions, the extent to which these benefits outweigh the safety costs is still uncertain. For instance, e-scooterists have been reported to experience a tenfold crash risk compared to ordinary cyclists. Today, we still do not know whether the real safety problem is the vehicle, the human, or the infrastructure. In other words, the new vehicles may not necessarily be unsafe; the behavior of their riders, in combination with an infrastructure that was not designed to accommodate micromobility, may be the real issue. Method: In this paper, we compared e-scooters and Segways with bicycles in field trials to determine whether these new vehicles create different constraints for longitudinal control (e.g., in braking avoidance maneuvers). Results: The results show that acceleration and deceleration performance changes across vehicles; specifically, e-scooters and Segways that we tested cannot brake as efficiently as bicycles. Further, bicycles are experienced as more stable, maneuverable, and safe than Segways and e-scooters. We also derived kinematic models for acceleration and braking that can be used to predict rider trajectories in active safety systems. Practical Applications: The results from this study suggest that, while new micromobility solutions may not be intrinsically unsafe, they may require some behavior and/or infrastructure adaptations to improve their safety. We also discuss how policy making, safety system design, and traffic education may use our results to support the safe integration of micromobility into the transport system. (c) 2022 The Author(s). Published by the National Safety Council and Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

When a cyclist's path intersects with that of a motorized vehicle at an unsignalized intersection, serious conflicts may happen. In recent years, the number of cyclist fatalities in this conflict scenario has held steady, while the number in many other traffic scenarios has been decreasing. There is, therefore, a need to further study this conflict scenario in order to make it safer. With the advent of automated vehicles, threat assessment algorithms able to predict cyclists' (other road users') behavior will be increasingly important to ensure safety. To date, the handful of studies that have modeled the vehicle-cyclist interaction at unsignalized intersections have used kinematics (speed and location) alone without using cyclists' behavioral cues, such as pedaling or gesturing. As a result, we do not know whether non-verbal communication (e.g., from behavioral cues) could improve model predictions. In this paper, we propose a quantitative model based on naturalistic data, which uses additional non-verbal information to predict cyclists' crossing intentions at unsignalized intersections. Interaction events were extracted from a trajectory dataset and enriched by adding cyclists' behavioral cues obtained from sensors. Both kinematics and cyclists' behavioral cues (e.g., pedaling and head movement), were found to be statistically significant for predicting the cyclist's yielding behavior. This research shows that adding information about the cyclists' behavioral cues to the threat assessment algorithms of active safety systems and automated vehicles will improve safety.

Introduction: Cycling is popular for its ecological, economic, and health benefits. However, especially in rural areas, cyclists may need to share the road with motorized traffic, which is often perceived as a threat. Overtaking a cyclist is a particularly critical maneuver for drivers as they need to control their lateral clearance and speed when passing the cyclist, possibly in the presence of oncoming vehicles or viewobstructing curves. An overtaking vehicle can destabilize the cyclist when passing with low clearance and high speed. At the same time, the cyclist may get scared and eventually stop cycling. In this work, we investigated how visibility regarding available sight distance-an important factor for infrastructure design and regulation-affects drivers' behavior when overtaking cyclists. Method: Using four roadsidebased traffic sensors, we collected naturalistic data that contained kinematics of drivers overtaking cyclists on a rural road in Sweden. We modeled lateral clearance and speed at the passing moment in response to variables such as sight distance and oncoming traffic with a Bayesian multivariate approach. Results: Fitted on 81 maneuvers, the model revealed that drivers reduced lateral clearance under reduced sight distance. Speed was similarly reduced, however, not as clearly. When an oncoming vehicle was present, it had a similar-yet stronger-effect than sight distance. While we found an overall correlation between clearance and speed, some maneuvers were recorded at critically low clearance. Conclusions: Cyclists' safety is endangered when passed by drivers under reduced visibility or close to oncoming traffic. Practical Applications: Decision-making for infrastructure and policymaking should aim at prohibiting overtaking in areas with reduced visibility or close oncoming traffic. The model developed in this study may serve as a reference to vehicle active-safety systems and automated driving. The collected and processed data may support evaluating driver models fitted on less ecologically valid data and simulated active-safety systems. (c) 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Existing research on control-transitions from automated driving (AD) to manual driving mainly stems from studies in virtual settings. There is a need for studies conducted in real settings to better understand the impacts of increasing vehicle automation on traffic safety. This study aims specifically to understand how drivers respond to take-over requests (TORs) in real traffic by investigating the associations between 1) where drivers look when receiving the TOR, 2) repeated exposure to TORs, and 3) the drivers’ response process. In total, thirty participants were exposed to four TORs after about 5–6 min of driving with AD on public roads. While in AD, participants could choose to engage in nondriving-related tasks (NDRTs). When they received the TOR, for 38% of TORs, participants were already looking on path. For those TORs where drivers looked off path at the time of the TOR, the off-path glance was most commonly towards an NDRT item. Then, for 72% of TORs (independent on gaze direction), drivers started their response process to the TOR by looking towards the instrument cluster before placing their hands on the steering wheel and their foot on the accelerator pedal, and deactivating automation. Both timing and order of these actions varied among participants, but all participants deactivated AD within 10 s from the TOR. The drivers’ gaze direction at the TOR had a stronger association with the response process than the repeated exposure to TORs did. Drivers can respond to TORs in real traffic. However, the response should be considered as a sequence of actions that requires a certain amount of time.

Recent advancements in technology make it possible for advanced driving assistance systems (ADAS) to recognize micromobility vehicles (MMV) and include them in their threat assessment. However, today we lack the rider-vehicle models which are of great importance in understanding the interconnection between the MMV and its rider. These models may help ADAS predict micromobility kinematics and provide accurate threat assessments, especially when avoidance maneuvers from micromobility must be considered. In this study, we modelled avoidance maneuvers from micromobility vehicles to support ADAS threat assessment. We compared traditional bicycles (with and without assistance) with e-scooters (a small personal scooter and a large scooter) in a field test, where 36 participants avoided a stationary obstacle by either braking or steering. Kinematic data such as longitudinal and latera speed, acceleration, jerk and steering angle and rate were collected and analyzed.

Introduction: In recent years, as novel micromobility vehicles (MMVs) have hit the market and rapidly gained popularity, new challenges in road safety have also arisen. There is an urgent need for validated models that comprehensively describe the behavior of such novel MMVs. This study aims to compare the longitudinal and lateral control of bicycles and e-scooters in a collision-avoidance scenario from a top down perspective, and to propose appropriate quantitative models for parameterizing and predicting the trajectories of the avoidance-braking and steering-maneuvers. Method: We compared a large e scooter and a light e-scooter with a bicycle (in assisted and non-assisted modes) in field trials to determine whether these new vehicles have different maneuverability constraints when avoiding a rear-end collision by braking and/or steering. Results: Braking performance in terms of deceleration and jerk varies among the different types of vehicles; specifically, e-scooters are not as effective at braking as bicycles, but the large e-scooter demonstrated better braking performance than the light one. No statistically significant difference was observed in the steering performance of the vehicles. Bicycles were perceived as more stable, maneuverable, and safe than e-scooters. The study also presents arctangent kinematic models for braking and steering, which demonstrate better accuracy and informativeness than linear models. Conclusions: This study demonstrates that the new micromobility solutions have some maneuverability characteristics that differ significantly from those of bicycles, and even within their own kind. Steering could be a more efficient collision-avoidance strategy for MMVs than braking under certain circumstances, such as in a rear-end collision. More complicated modeling for MMV kinematics can be beneficial but needs validation. Practical Applications: The proposed arctangent models could be used in new advanced driving assistance systems to prevent crashes between cars and MMV users. Micromobility safety could be improved by educating MMV riders to adapt their behavior accordingly. Further, knowledge about the differences in maneuverability between e-scooters and bicycles could inform infrastructure design, and traffic regulations.(c) 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Drivers overtaking cyclists on rural roads are a safety concern, as drivers need to handle the interaction with the cyclist and possibly an oncoming vehicle. Improving the maneuver’s outcome requires an understanding of not only the objective, measurable safety metrics, but also the subjective, perceived safety of each road user. Previous research has shown that the perceived safety of the cyclist is most at risk at the passing moment, when driver and cyclist are closest to each other. However, to develop safety measures, it is necessary to know how both road users perceive safety, by understanding the factors that influence their perceptions during the overtaking maneuver. This study measured the perceived safety of drivers in a test-track experiment in Sweden and the perceived safety of cyclists in a field test in Spain. For both drivers and cyclists, we developed Bayesian ordinal logistic regression models of perceived safety scores that take as input objective safety metrics representing the different crash risks at the passing moment. Our results show that while drivers’ perceived safety decreases when there is an oncoming vehicle with a low time-to-collision, cyclists’ perceived safety is reduced by a small lateral clearance and a high overtaking speed. Although our datasets are heterogeneous and limited, our results are in line with previous research. In addition, the Bayesian models presented in this paper are novel and may be improved in future studies once more naturalistic data become available. We discuss how our models may support infrastructure development and regulation, policymaking, driver coaching, the development of active safety systems, and automated driving by providing a possible method for predicting perceived safety.

The total number of road crashes in Europe is decreasing, but the number of crashes involving cyclists is not decreasing at the same rate. When cars and bicycles share the same lane, cars typically need to overtake them, creating dangerous conflicts-especially on rural roads, where cars travel much faster than cyclists. In order to protect cyclists, advanced driver assistance systems (ADAS) are being developed and introduced to the market. One of them is a forward collision warning (FCW) system that helps prevent rear-end crashes by identifying and alerting drivers of threats ahead. The objective of this study is to assess the relative safety benefit of a behaviour-based (BB) FCW system that protects cyclists in a car-to-cyclist overtaking scenario. Virtual safety assessments were performed on crashes derived from naturalistic driving data. A series of driver response models was used to simulate different driver reactions to the warning. Crash frequency in conjunction with an injury risk model was used to estimate the risk of cyclist injury and fatality. The virtual safety assessment estimated that, compared to no FCW, the BB FCW could reduce cyclists' fatalities by 53-96% and serious injuries by 43-94%, depending on the driver response model. The shorter the driver's reaction time and the larger the driver's deceleration, the greater the benefits of the FCW. The BB FCW also proved to be more effective than a reference FCW based on the Euro NCAP standard test protocol. The findings of this study demonstrate the BB FCW's great potential to avoid crashes and reduce injuries in car-to-cyclist overtaking scenarios, even when the driver response model did not exceed a comfortable rate of deceleration. The results suggest that a driver behaviour model integrated into ADAS collision threat algorithms can provide substantial safety benefits.