Get Ready for Take-Overs: Using Head-Up Display for Drivers to Engage in Non-Driving-Related Tasks in Automated Vehicles.

OBJECTIVE: The study aims to investigate the potential of using HUD (head-up display) as an approach for drivers to engage in non-driving-related tasks (NDRTs) during automated driving, and examine the impacts on driver state and take-over performance in comparison to the traditional mobile phone. BACKGROUND: Advances in automated vehicle technology have the potential to relieve drivers from driving tasks so that they can engage in NDRTs freely. However, drivers will still need to take-over control under certain circumstances. METHOD: A driving simulation experiment was conducted using an Advanced Driving Simulator and real-world driving videos. Forty-six participants completed three drives in three display conditions, respectively (HUD, mobile phone and baseline without NDRT). The HUD was integrated with the vehicle in displaying NDRTs while the mobile phone was not. Drivers' visual (e.g. gaze, blink) and physiological (e.g. ECG, EDA) data were collected to measure driver state. Two take-over reaction times (hand and foot) were used to measure take-over performance. RESULTS: The HUD significantly shortened the take-over reaction times compared to the mobile phone condition. Compared to the baseline condition, drivers in the HUD condition also experienced lower cognitive workload and physiological arousal. Drivers' take-over reaction times were significantly correlated with their visual and electrodermal activities during automated driving prior to the take-over request. CONCLUSION: HUDs can improve driver performance and lower workload when used as an NDRT interface. APPLICATION: The study sheds light on a promising approach for drivers to engage in NDRTs in future AVs.

Is my AV crashing? An online photo-based experiment assessing whether shared intended pathway can help AV drivers anticipate silent failures.

The shared responsibility between conditional AVs drivers demands shared understanding. Thus, a shared intended pathway (SIP)-a graphical display of the AV's planned manoeuvres in a head-up display to help drivers anticipate silent failures is proposed. An online, randomised photo experiment was conducted with 394 drivers in Australia. The photos presented traffic scenarios where the SIP forecast either safe or unsafe manoeuvres (silent failures). Participants were required to respond by selecting whether driver intervention was necessary or not. Additionally, the effects of presented object recognition bounding boxes which indicated whether a road user was recognised or not were also tested in the experiment. The SIP led to correct intervention choices 87% of the time, and to calibrating self-reported trust, perceived ease of use and usefulness. The bounding boxes found no significant effects. Results suggest SIPs can assist in monitoring conditional automation. Future research in simulator studies is recommended. Practitioner summary: Conditional AV drivers are expected to take-over control during failures. However, drivers are not informed about the AV's planned manoeuvres. A visual display that presents the shared intended pathway is proposed to help drivers mitigate silent failures. This online photo experiment found the display helped anticipate failures with 87% accuracy.

A Novel Crowdsourcing Model for Micro-Mobility Ride-Sharing Systems.

Substantial research is required to ensure that micro-mobility ride sharing provides a better fulfilment of user needs. This study proposes a novel crowdsourcing model for the ride-sharing system where light vehicles such as scooters and bikes are crowdsourced. The proposed model is expected to solve the problem of charging and maintaining a large number of light vehicles where these efforts will be the responsibility of the crowd of suppliers. The proposed model consists of three entities: suppliers, customers, and a management party responsible for receiving, renting, booking, and demand matching with offered resources. It can allow suppliers to define the location of their private e-scooters/e-bikes and the period of time they are available for rent. Using a dataset of over 9 million e-scooter trips in Austin, Texas, we ran an agent-based simulation six times using three maximum battery ranges (i.e., 35, 45, and 60 km) and different numbers of e-scooters (e.g., 50 and 100) at each origin. Computational results show that the proposed model is promising and might be advantageous to shift the charging and maintenance efforts to a crowd of suppliers.

Validation of a Driving Simulator Study on Driver Behavior at Passive Rail Level Crossings.

OBJECTIVE: The behavioral validation of an advanced driving simulator for its use in evaluating passive level crossing countermeasures was performed for stopping compliance and speed profile. BACKGROUND: Despite the fact that most research on emerging interventions for improving level crossing safety is conducted in a driving simulator, no study has validated the use of a simulator for this type of research. METHOD: We monitored driver behavior at a selected passive level crossing in the Brisbane region in Australia for 3 months ( N = 916). The level crossing was then replicated in an advanced driving simulator, and we familiarized participant drivers ( N = 54) with traversing this crossing, characterized by low road and rail traffic. RESULTS: We established relative validity for the stopping compliance and the approach speed. CONCLUSION: This validation study suggests that driving simulators are an appropriate tool to study the effects of interventions at passive level crossing with low road and rail traffic, which are prone to reduced compliance due to familiarity. APPLICATION: This study also provides support for the findings of previous driving simulator studies conducted to evaluate compliance and approach speeds of passive level crossings.

The 'invisible gorilla' during pedestrian-AV interaction: Effects of secondary tasks on pedestrians' reaction to eHMIs.

In road traffic, mental overload often leads to a failure to notice new and distinctive stimuli. Such phenomenon is known as 'inattentional blindness'. Safe and efficient interaction between automated vehicles (AVs) and pedestrians is expected to rely heavily on external human-machine interfaces (eHMIs), a tool AVs are equipped with to communicate their intentions to pedestrians. This study seeks to explore the phenomenon of 'inattentional blindness' in the context of pedestrian-AV interactions. Specifically, the aim is to understand the effects of a warning eHMI on pedestrians' crossing decisions when they are engaged in a secondary task. In an experiment study with videos of pedestrian crossing scenarios filmed from the perspective of the crossing pedestrian, participants had to decide the latest point at which they would be willing to cross the road in front of an AV with an eHMI vs. an AV without an eHMI. Participants were also asked to predict the future behavior of the AV. 125 female and 9 male participants aged between 18 and 25 completed the experiment and a follow-up questionnaire. It was found that the presence of a warning eHMI on AVs contributes to a clearer understanding of pedestrians' inferences about the intention of AVs and helps deter late and dangerous crossing decisions made by pedestrians. However, the eHMI fail to help pedestrians avoid such decisions when they face a high mental workload induced by secondary task engagement.

Crash severity analysis and risk factors identification based on an alternate data source: a case study of developing country.

Road traffic injuries are one of the primary reasons for death, especially in developing countries like Bangladesh. Safety in land transport is one of the major concerns for road safety authorities and other policymakers. For this reason, contributory factors identification associated with crashes is necessary for reducing road crashes and ensuring transportation safety. This paper presents an analytical approach to identifying significant contributing factors of Bangladesh road crashes by evaluating the road crash data, considering three different severity levels (non-fetal, severe, and extremely severe). Generally, official crash databases are compiled from police-reported crash records. Though the official datasets are focusing on compiling a wide array of attributes, an assorted number of unreported issues can be observed that demands an alternative source of crash data. Therefore, this proposed approach considers compiling crash data from newspapers in Bangladesh which could be complimentary to the official crash database. To conduct the analysis, first, we filtered the useful features from compiled crash data using three popular feature selection techniques: chi-square, Two-way ANOVA, and Regression analysis. Then, we employed three machine learning classifiers: Decision Tree, Random Forest, and Naïve Bayes over the extracted features. A confusion matrix was considered to evaluate the proposed model, including classification accuracy, sensitivity, and specificity. The predictive machine learning model, namely, Random Forest using Label Encoder with chi-square and Two-way ANOVA feature selection process, seems the best option for crash severity prediction that provides high prediction accuracy. The resulting model highlights nine out of fourteen independent features as responsible factors. Significant features associated with crash severities include driver characteristics (gender, license type, seat belts), vehicle characteristics (vehicle type), road characteristics (road surface type, road classification), environmental conditions (day of crash occurred, time of crash), and injury localization. This outcome may contribute to improving traffic safety of Bangladesh.

Crash severity analysis of vulnerable road users using machine learning.

Road crash fatality is a universal problem of the transportation system. A massive death toll caused annually due to road crash incidents, and among them, vulnerable road users (VRU) are endangered with high crash severity. This paper focuses on employing machine learning-based classification approaches for modelling injury severity of vulnerable road users-pedestrian, bicyclist, and motorcyclist. Specifically, this study aims to analyse critical features associated with different VRU groups-for pedestrian, bicyclist, motorcyclist and all VRU groups together. The critical factor of crash severity outcomes for these VRU groups is estimated in identifying the similarities and differences across different important features associated with different VRU groups. The crash data for the study is sourced from the state of Queensland in Australia for the years 2013 through 2019. The supervised machine learning algorithms considered for the empirical analysis includes the K-Nearest Neighbour (KNN), Support Vector Machine (SVM) and Random Forest (RF). In these models, 17 distinct road crash parameters are considered as input features to train models, which originate from road user characteristics, weather and environment, vehicle and driver condition, period, road characteristics and regions, traffic, and speed jurisdiction. These classification models are separately trained and tested for individual and unified VRU to assess crash severity levels. Afterwards, model performances are compared with each other to justify the best classifier where Random Forest classification models for all VRU modes are found to be comparatively robust in test accuracy: (motorcyclist: 72.30%, bicyclist: 64.45%, pedestrian: 67.23%, unified VRU: 68.57%). Based on the Random Forest model, the road crash features are ranked and compared according to their impact on crash severity classification. Furthermore, a model-based partial dependency of each road crash parameters on the severity levels is plotted and compared for each individual and unified VRU. This clarifies the tendency of road crash parameters to vary with different VRU crash severity. Based on the outcome of the comparative analysis, motorcyclists are found to be more likely exposed to higher crash severity, followed by pedestrians and bicyclists.

Driver influence on vehicle trajectory prediction.

Drivers continually interact with other road users and use information from the road environment to make decisions to control their vehicle. A clear understanding of different parameters impacting this interaction can provide us with a new design approach for a more effective driver assistance system - a personalised trajectory prediction system. This paper highlights the influential factors on trajectory prediction system performance by (i) identifying driver behaviours impacting the trajectory prediction system; and (ii) analysing other contributing factors such as traffic density, secondary task, gender and age group. To explore the most influential contributing factors, we first train an interaction-aware trajectory prediction system using time-series data derived from the Second Strategic Highway Research Program (SHRP2) Naturalistic Driving Study (NDS). Prediction error is then analysed based on driver characteristics such as driver profile which is subjectively measured through self-reported questions, and driving performance which is based on evaluation of time-series information such as speed, acceleration, jerk, time, and space headway. The results show that prediction error significantly increased in the scenarios where the driver engaged in risky behaviour. Analysis shows that trajectory prediction system performance is also affected by factors such as traffic density, engagement in secondary tasks, driver gender and age group. We show that the driver profile, which is subjectively measured using self-reported questionnaires, is not as significant as the driving performance information, which is objectively measured and extracted during each specific driving scenario.

Correction: A framework for testing independence between lane change and cooperative intelligent transportation system.

[This corrects the article DOI: 10.1371/journal.pone.0229289.].

Autonomous Vehicles and Vulnerable Road-Users-Important Considerations and Requirements Based on Crash Data from Two Countries.

(1) Background: Passenger vehicles equipped with advanced driver-assistance system (ADAS) functionalities are becoming more prevalent within vehicle fleets. However, the full effects of offering such systems, which may allow for drivers to become less than 100% engaged with the task of driving, may have detrimental impacts on other road-users, particularly vulnerable road-users, for a variety of reasons. (2) Crash data were analysed in two countries (Great Britain and Australia) to examine some challenging traffic scenarios that are prevalent in both countries and represent scenarios in which future connected and autonomous vehicles may be challenged in terms of safe manoeuvring. (3) Road intersections are currently very common locations for vulnerable road-user accidents; traffic flows and road-user behaviours at intersections can be unpredictable, with many vehicles behaving inconsistently (e.g., red-light running and failure to stop or give way), and many vulnerable road-users taking unforeseen risks. (4) Conclusions: The challenges of unpredictable vulnerable road-user behaviour at intersections (including road-users violating traffic or safe-crossing signals, or taking other risks) combined with the lack of knowledge of CAV responses to intersection rules, could be problematic. This could be further compounded by changes to nonverbal communication that currently exist between road-users, which could become more challenging once CAVs become more widespread.

Hybrid pointer networks for traveling salesman problems optimization.

In this work, we proposed a hybrid pointer network (HPN), an end-to-end deep reinforcement learning architecture is provided to tackle the travelling salesman problem (TSP). HPN builds upon graph pointer networks, an extension of pointer networks with an additional graph embedding layer. HPN combines the graph embedding layer with the transformer's encoder to produce multiple embeddings for the feature context. We conducted extensive experimental work to compare HPN and Graph pointer network (GPN). For the sack of fairness, we used the same setting as proposed in GPN paper. The experimental results show that our network significantly outperforms the original graph pointer network for small and large-scale problems. For example, it reduced the cost for travelling salesman problems with 50 cities/nodes (TSP50) from 5.959 to 5.706 without utilizing 2opt. Moreover, we solved benchmark instances of variable sizes using HPN and GPN. The cost of the solutions and the testing times are compared using Linear mixed effect models. We found that our model yields statistically significant better solutions in terms of the total trip cost. We make our data, models, and code publicly available https://github.com/AhmedStohy/Hybrid-Pointer-Networks.

Real-time performance modelling of a Sustained Attention to Response Task.

Vigilance declines when exposed to highly predictable and uneventful tasks. Monotonous tasks provide little cognitive and motor stimulation and contribute to human errors. This paper aims to model and detect vigilance decline in real time through participants' reaction times during a monotonous task. A laboratory-based experiment adapting the Sustained Attention to Response Task (SART) is conducted to quantify the effect of monotony on overall performance. Relevant parameters are then used to build a model detecting hypovigilance throughout the experiment. The accuracy of different mathematical models is compared to detect in real time - minute by minute - the lapses in vigilance during the task. It is shown that monotonous tasks can lead to an average decline in performance of 45%. Furthermore, vigilance modelling enables the detection of vigilance decline through reaction times with an accuracy of 72% and a 29% false alarm rate. Bayesian models are identified as a better model to detect lapses in vigilance as compared with neural networks and generalised linear mixed models. This modelling could be used as a framework to detect vigilance decline of any human performing monotonous tasks. STATEMENT OF RELEVANCE: Existing research on monotony is largely entangled with endogenous factors such as sleep deprivation, fatigue and circadian rhythm. This paper uses a Bayesian model to assess the effects of a monotonous task on vigilance in real time. It is shown that the negative effects of monotony on the ability to sustain attention can be mathematically modelled and predicted in real time using surrogate measures, such as reaction times. This allows the modelling of vigilance fluctuations.

Drivers' gap acceptance behaviours at intersections: A driving simulator study to understand the impact of mobile phone visual-manual interactions.

Mobile phone use is often considered to be the main source of distraction on the road. Gap acceptance at intersections is a frequent and complex driving task that requires high visual attention from drivers. This study aims to investigate the effect of mobile phone use on the gap acceptance manoeuvre at intersections. Different mobile phone use positions, intersection type, gap size and driver characteristics were considered in the study. A total of 41 licenced drivers drove in an advanced driving simulator in three phone use conditions: baseline (no phone use), using the phone under the steering wheel (covert) and using the phone above the steering wheel (overt). Drivers drove the simulator three times and experienced two intersection types (straight-forward vs. left-turn) and two gap sizes (4 s vs. 7 s) during each drive. A parametric accelerated failure time (AFT) duration model was developed to evaluate the intersection crossing completion time of drivers. The results showed no significant difference of gap acceptance behaviours between the two phone use positions. The distraction task did not affect drivers' gap acceptance decision, but it increased the crossing completion time by over 10 % compared to baseline. Besides, drivers behaved conservatively at intersections while using a mobile phone, such as adopting a larger deceleration, waiting a longer time, and mainting a larger distance to the front vehicle, etc. However, these compensational behaviours were not helpful in improving the intersection traffic situation regarding both safety and efficiency. Intersection type and gap size were both significant factors of gap acceptance decision and crossing completion time. Additionally, younger drivers were more likely to accept a gap than older drivers, and female drivers spent longer time to cross the intersection than males.

Exploring drivers' mental workload and visual demand while using an in-vehicle HMI for eco-safe driving.

Eco-safe driving is a promising approach to improve road safety while reducing transport emissions. The application of an eco-safe driving system is feasible with the support of vehicle-to-vehicle/infrastructure technologies. To guarantee system usability and safety appropriateness, a key precondition is to ensure that driver mental workload and visual demands required for using the system are reasonable. This study explored how drivers' mental workload and visual demands were affected when driving with an eco-safe driving HMI (human-machine-interface). Four in-vehicle eco-safe HMI information conditions were evaluated, including baseline, advice only, feedback only, and advice & feedback. Two traffic scenarios (stop-sign intersection with traffic vs. stop-sign intersection without traffic) were simulated using an advanced driving simulator. Behavioural variables (e.g. brake force, acceleration), visual variables (e.g. blink metrics, pupil size) and subjective workload scores were collected from 36 licensed Australian drivers. The experiment results showed that the HMI prompted drivers to apply a smooth and stable brake force when they approached the intersection and a smooth acceleration when they left the intersection. Drivers' mental workload indicated by visual measurements were consistent with their subjective reported workload levels. Drivers had a higher mental workload when they received and processed additional eco-safe information in the advice & feedback condition. An increase in mental workload induced by the in-vehicle cognitive task initiated more blink activities while the increase in visual demand caused by a complex road situation led to blink inhibition. The study shows the HMI could significantly promote eco-safe driving behaivours without causing excessive mental and visual workload of drivers.

A framework for testing independence between lane change and cooperative intelligent transportation system.

Cooperative Intelligent Transportation Systems (C-ITS) are being deployed in several cities around the world. We are preparing for the largest Field Operational Test (FOT) in Australia to evaluate C-ITS safety benefits. Two of the safety benefit hypotheses we formulated assume a dependency between lane changes and C-ITS warnings displayed on the Human Machine Interface (HMI) during safety events. Lane change detection is done by processing many predictors from several sensors at the time of the safety event. However, in our planned FOT, the participating vehicles are only equipped with the vehicle C-ITS and the IMU. Therefore, in this paper, we propose a framework to test lane change and C-ITS dependency. In this framework, we train a random forest classifier using data collected from the IMU to detect lane changes. Consequently, the random forest output probabilities of the testing data in case of C-ITS and control are used to construct a 2x2 contingency table. Then we develop a permutation test to calculate the null hypothesis needed to test the independence of the lane change during safety events and the C-ITS.

How do drivers avoid collisions? A driving simulator-based study.

INTRODUCTION: Drivers' collision avoidance performance in an impending collision situation plays a decisive role for safety outcomes. This study explored drivers' collision avoidance performances in three typical collision scenarios that were right-angle collision, head-on collision, and collision with pedestrian. METHOD: A high-fidelity driving simulator was used to design the scenarios and conduct the experiment. 45 participants took part in the simulator experiment. Drivers' longitudinal/lateral collision avoidance performances and collision result were recorded. RESULTS: Experimental results showed that brake only was the most common response among the three collision scenarios, followed by brake combining swerve in head-on and pedestrian collision scenarios. In right-angle collision scenario with TTC (time to collision) largest among three scenarios, no driver swerved, and meanwhile drivers who showed slow brake reaction tended to compensate the collision risk by taking a larger maximum deceleration rate within a shorter time. Swerve-toward-conflict was a prevalent phenomenon in head-on and pedestrian collision scenarios and significantly associated with collision risk. Drivers that swerved toward the conflict object had a shorter swerve reaction time than drivers that swerved away from conflict. CONCLUSIONS: Long brake reaction time and wrong swerve direction were the main factors leading to a high collision likelihood. The swerve-toward-conflict maneuver caused a delay in brake action and degraded subsequent braking performances. The prevalent phenomenon indicated that drivers tended to use an intuitive (heuristic) way to make decisions in critical traffic situations. Practical applications: The study generated a better understanding of collision development and shed lights on the design of future advanced collision avoidance systems for semi-automated vehicles. Manufactures should also engage more efforts in developing active steering assistance systems to assist drivers in collision avoidance.

Evaluation of in-vehicle technologies to prevent unlicensed driving in Queensland and Victoria.

Unlicensed driving is a serious problem in many Australian states, with unlicensed driving-related crashes (UDC) costing up to $304 million per year in Queensland, and $176 million in Victoria. In this paper, we present a Benefit-Cost Ratio (BCR) analysis of a set of Intelligent Transportation Systems technologies aimed at preventing unlicensed driving by verifying the driver's identity through biometric technology, as well as the validity of their licence. Utilised together, the technology would essentially take the form of a licence interlock. The goal of this program of research (from which this paper stems) was to provide preliminary recommendations as to which technology is the most beneficial and should be implemented as part of a government-led program increasing the functionalities of electronic driving licences (EDL). The corresponding BCR analysis revealed that fingerprints and finger vascular patterns recognition technologies were found to systematically have the best BCRs. In regard to the most effective manner to implement the technology, a corresponding investigation with five scenarios revealed that the greatest benefits would be achieved with: (a) a mandatory system for all banned drivers (e.g., suspensions & disqualifications), and (b) a mandatory system for banned drivers under the age of 21 only. Scenario (b) performs extremely well, with returns of up to 16 times the investment with a simple fingerprint-based interlock. Although often more modest, all systems were found to have BCRs above 1 in all of the implementation scenarios except one. This paper further outlines the findings in regard to addressing the significant problem of unlicensed driving via emerging technologies.

Effects of an in-vehicle eco-safe driving system on drivers' glance behaviour.

We have designed a new in-vehicle eco-safe driving system and shown its effectiveness in prompting drivers to execute a fuel-saving and safe driving style (Vaezipour et al., 2018, submitted for publication). However, the system could also bring potential negative outcomes, i.e. driver distraction. This simulator study investigated drivers' glance behaviours as indicators of driver distraction when using our Eco-Safe Human-Machine-Interface (HMI). Four types of eco-safe information display conditions (baseline, advice only, feedback only, both advice and feedback) were tested on different traffic situations with varied road traffic complexity. Results showed that the eco-safe HMI system did not cause visual distraction. In contrast, the advice only or feedback only information improved forward gazing on the roadway. In addition, drivers tended to adapt their visual scanning strategies according to the traffic situations. In the car-following situation they paid longer glances to the forward roadway, while in the intersections they spent more time to look at the HMI system. The findings indicated that our eco-safe driving system improved drivers' eco-safe behaviours and meanwhile enhanced their visual attention on road while no evidence showed that drivers were distracted by it.

Mistakes or deliberate violations? A study into the origins of rule breaking at pedestrian train crossings.

Train pedestrian collisions are the most likely to result in severe injuries and fatalities when compared to other types of rail crossing accidents. However, there is currently scant research that has examined the origins of pedestrians' rule breaking at level crossings. As a result, this study examined the origins of pedestrians' rule breaking behaviour at crossings, with particular emphasis directed towards examining the factors associated with making errors versus deliberation violations. A total of 636 individuals volunteered to participate in the study and completed either an online or paper version of the questionnaire. Quantitative analysis of the data revealed that knowledge regarding crossing rules was high, although up to 18% of level crossing users were either unsure or did not know (in some circumstances) when it was legal to cross at a level crossing. Furthermore, 156 participants (24.52%) reported having intentionally violated the rules at level crossings and 3.46% (n=22) of the sample had previously made a mistake at a crossing. In regards to rule violators, males (particularly minors) were more likely to report breaking rules, and the most frequent occurrence was after the train had passed rather than before it arrives. Regression analysis revealed that males who frequently use pedestrian crossings and report higher sensation seeking traits are most likely to break the rules. This research provides evidence that pedestrians are more likely to deliberately violate rules (rather than make errors) at crossings and it illuminates high risk groups. This paper will further outline the study findings in regards to the development of countermeasures as well as provide direction for future research efforts in this area.

Driver's behavioural changes with new intelligent transport system interventions at railway level crossings--A driving simulator study.

Improving safety at railway level crossings is an important issue for the Australian transport system. Governments, the rail industry and road organisations have tried a variety of countermeasures for many years to improve railway level crossing safety. New types of intelligent transport system (ITS) interventions are now emerging due to the availability and the affordability of technology. These interventions target both actively and passively protected railway level crossings and attempt to address drivers' errors at railway crossings, which are mainly a failure to detect the crossing or the train and misjudgement of the train approach speed and distance. This study aims to assess the effectiveness of three emerging ITS that the rail industry considers implementing in Australia: a visual in-vehicle ITS, an audio in-vehicle ITS, as well as an on-road flashing beacons intervention. The evaluation was conducted on an advanced driving simulator with 20 participants per trialled technology, each participant driving once without any technology and once with one of the ITS interventions. Every participant drove through a range of active and passive crossings with and without trains approaching. Their speed approach of the crossing, head movements and stopping compliance were measured. Results showed that driver behaviour was changed with the three ITS interventions at passive crossings, while limited effects were found at active crossings, even with reduced visibility. The on-road intervention trialled was unsuccessful in improving driver behaviour; the audio and visual ITS improved driver behaviour when a train was approaching. A trend toward worsening driver behaviour with the visual ITS was observed when no trains were approaching. This trend was not observed for the audio ITS intervention, which appears to be the ITS intervention with the highest potential for improving safety at passive crossings.