Detecting abnormal driving behavior is critical for road traffic safety and the evaluation of drivers’ behavior. With the advancement of machine learning (ML) algorithms and the accumulation of naturalistic driving data, many ML models have been adopted for abnormal driving behavior detection (also referred to in this paper as “anomalies”). Most existing ML-based detectors rely on (fully) supervised ML methods, which require substantial labeled data. However, ground truth labels are not always available in the real world, and labeling large amounts of data is tedious. Thus, there is a need to explore unsupervised or semi-supervised methods to make the anomaly detection process more feasible and efficient. To fill this research gap, this study analyzes large-scale real-world data revealing several abnormal driving behaviors (e.g., sudden acceleration, rapid lane-changing) and develops a hierarchical extreme learning machine (HELM)-based semi-supervised ML method using partly labeled data to accurately detect the identified abnormal driving behaviors. Moreover, previous ML-based approaches predominantly utilized basic vehicle motion features (such as velocity and acceleration) to label and detect abnormal driving behaviors, while this study seeks to introduce event-level safety indicators as input features for ML models to improve the detection performance. Results from extensive experiments demonstrate the effectiveness of the proposed semi-supervised ML model with the introduced safety indicators serving as important features. The proposed semi-supervised ML method outperforms other baseline semi-supervised or unsupervised methods as far as various metrics are concerned: for example, it delivers the best accuracy at 99.58% and the best F1-score at 0.9913. The ablation study further highlights the significance of safety indicators for advancing the detection performance of abnormal driving behaviors.

As automated vehicles (AVs) become increasingly popular, the question arises as to how cyclists will interact with such vehicles. This study investigated (1) whether cyclists spontaneously notice if a vehicle is driverless, (2) how well they perform a driver-detection task when explicitly instructed, and (3) how they carry out these tasks. Using a Wizard-of-Oz method, 37 participants cycled a designated route and encountered an AV multiple times in two experimental sessions. In Session 1, participants cycled the route uninstructed, while in Session 2, they were instructed to verbally report whether they detected the presence or absence of a driver. Additionally, we recorded participants’ gaze behaviour with eye-tracking and their responses in post-session interviews. The interviews revealed that 30% of the cyclists spontaneously mentioned the absence of a driver (Session 1), and when instructed (Session 2), they detected the absence and presence of the driver with 93% accuracy. The eye-tracking data showed that cyclists looked more frequently and for longer at the vehicle in Session 2 compared to Session 1. Additionally, participants exhibited intermittent sampling of the vehicle, and they looked at the area in front of the vehicle when it was far away and towards the windshield region when it was closer. The post-session interviews also indicated that participants were curious, but felt safe, and reported a need to receive information about the AV's driving state. In conclusion, cyclists can detect the absence of a driver in the AV, and this detection may influence their perception of safety. Further research is needed to explore these findings in real-world traffic conditions.

As automated vehicles (AVs) become more common, it is important to understand how human-driven vehicles (HDVs) would interact with them. This research investigated HDV gap acceptance behavior in mixed traffic with AVs at a priority intersection, focusing on how mixed traffic factors affect this behavior and overall traffic efficiency. Using a driving simulator, four scenarios were tested by varying AV driving style (less defensive, more defensive, and HDV-like) and AV recognizability (distinguishable or not from HDVs). Gap acceptance models were estimated based on the collected trajectory data. These models were then integrated into the SUMO microscopic traffic simulation platform, where a T-intersection network was set up. Simulation runs varied based on AV driving style, recognizability, penetration rate (0-75% in 25% increments), and whether HDV behavioral adaptation was considered. The results indicated increased minor road vehicle delays with higher AV penetration rates. Recognizable less defensive AVs, and more defensive AVs with high penetration rates caused the largest delays for minor road vehicles compared to other conditions. Ignoring behavioral adaptation led to a delay underestimation of up to 75% for minor road vehicles. In conclusion, there is behavioral adaptation in gap acceptance of HDVs in mixed traffic environments. Taking into account the behavioral adaptation is essential for accurately assessing traffic efficiency in mixed traffic conditions, and guiding AV deployment policies.

The burgeoning navigation services using digital maps provide great convenience to drivers. Nevertheless, the presence of anomalies in lane-rendering map images occasionally introduces potential hazards, as such anomalies can mislead human drivers and consequently contribute to unsafe driving. In response to this concern to accurately and effectively detect the anomalies, this paper transforms lane-rendering image anomaly detection into a classification problem and proposes a four-phase pipeline: data preprocessing, self-supervised pretraining with the masked image modeling (MiM) method, customized fine-tuning using cross-entropy-based loss with label smoothing, and post-processing. Leveraging state-of-the-art deep learning techniques, especially those involving transformer models, the pipeline demonstrates superior performance verified through various experiments. Notably, self-supervised pretraining with MiM can greatly enhance detection accuracy while significantly reducing the total training time. For instance, employing the Swin Transformer with Uniform Masking as self-supervised pretraining yielded a higher accuracy of 94.77% and an improved area under the curve (AUC) score of 0.9743 compared with the pure Swin Transformer without pretraining with an accuracy of 94.01% and an AUC of 0.9498. Furthermore, fine-tuning epochs were dramatically reduced to 41 from the original 280. Ablation study with regard to techniques to alleviate the data imbalance between normal and abnormal instances further reinforces the model’s overall performance. In conclusion, the proposed pipeline, with its incorporation of self-supervised pretraining using MiM and other advanced deep learning techniques, emerges as a robust solution for enhancing the accuracy and efficiency of lane-rendering image anomaly detection in digital navigation systems.

Dedicated Lanes (DLs) have been proposed as a potential alternative for the deployment of Connected and Automated Vehicles (CAVs) to facilitate platooning and increase motorway capacity. However, the impact of the presence and utilization policy of such a lane on drivers’ preference to use automation and their behaviour has not yet been thoroughly investigated. In this study, a driving simulator experiment is conducted, where participants drive a CAV in the presence of a DL with different utilization policies. Drivers have the possibility to choose between driving in an automated mode or in a manual mode. In automated mode they could adjust the driving speed and time headway and initiate automated lane changes. Two utilization policies were examined: mandatory versus optional use of DLs when driving in an automated mode. The impact of the presence and utilization policy of the DL on drivers’ preference to use automation and their behaviour in car-following and lane changing are investigated. The study found that while the presence of a DL does not increase drivers’ preference for automation use, it encourages drivers to utilize the DL more when the utilization policy is mandatory (i.e., drivers can only use automation mode when driving on this lane). Furthermore, drivers are more conservative in automated mode and when driving in mixed traffic. However, they perform closer car-following and merge into smaller gaps when driving on DLs which on one hand can increase the capacity of the DLs, but on the other hand can increase the risk of collisions. These results are useful for road operators, and in setting-up a more realistically traffic simulation studies.

Understanding human behavior in overtaking scenarios is crucial for enhancing road safety in mixed traffic with automated vehicles (AVs). Computational models of behavior play a pivotal role in advancing this understanding, as they can provide insight into human behavior generalizing beyond empirical studies. However, existing studies and models of human overtaking behavior have mostly focused on scenarios with simplistic, constant-speed dynamics of oncoming vehicles, disregarding the potential of AVs to proactively influence the decision-making process of the human drivers via implicit communication. Furthermore, despite numerous studies in other scenarios, so far it remained unknown whether overtaking decisions of human drivers are affected by whether they are interacting with an AV or a human-driven vehicle (HDV). To address these gaps, we conducted a “reverse Wizard-of-Oz” driving simulator experiment with 30 participants who repeatedly interacted with oncoming AVs and HDVs, measuring the drivers' gap acceptance decisions and response times. The oncoming vehicles featured time-varying dynamics designed to influence the overtaking decisions of the participants by briefly decelerating and then recovering to their initial speed. We found no evidence of differences in participants' overtaking behavior when interacting with oncoming AVs compared to HDVs. Furthermore, we did not find any evidence of brief decelerations of the oncoming vehicle affecting the decisions or response times of the participants. Cognitive modeling of the obtained data revealed that a generalized drift-diffusion model with dynamic drift rate and velocity-dependent decision bias best explained the gap acceptance outcomes and response times observed in the experiment. Overall, our findings highlight that cognitive models of the kind considered here can provide a generalizable description of human overtaking decisions and their timing. Such models can thus help further advance the ongoing development of safer interactions between human drivers and AVs during overtaking maneuvers.

The burgeoning navigation services using digital maps provide great convenience to drivers. However, there are sometimes anomalies in the lane rendering map images, which might mislead human drivers and result in unsafe driving. To accurately and effectively detect the anomalies, this paper transforms lane rendering image anomaly detection into a classification problem and proposes a four-phase pipeline consisting of data pre-processing, self-supervised pre-training with the masked image modeling (MiM) method, customized fine-tuning using cross-entropy loss with label smoothing, and post-processing to tackle it using state-of-the-art deep learning techniques, especially the Transformer models. Various experiments verify the effectiveness of the proposed pipeline. The proposed pipeline can deliver superior lane rendering image anomaly detection performance, and especially, the self-supervised pre-training with MiM can greatly improve the detection accuracy while significantly reducing the total training time, e.g, Swin Transformer with Uniform Masking as self-supervised pretraining (Swin-Trans-UM) obtained better accuracy at 94.77% and better Area Under The Curve (AUC) at 0.9743 compared with the pure Swin Transformer without pre-training (Swin-Trans) whose accuracy is 94.01% AUC is 0.9498, and the fine-tuning epochs reduced to 41 from original 280. Ablation study further regarding techniques to alleviate the data imbalance between normal and abnormal instances further enhances the model performance.

Sharp curves in freeways are known to be unsafe design elements since drivers do not expect them. It is difficult for drivers to estimate the radius of a curve. Therefore, drivers are believed to use other cues to decelerate when approaching a curve. Based on previous successful experiences of driven speeds in curves, drivers are thought to have built expectations of safe speeds given certain cues, minimalising risks. This research employs a Bayesian Belief Network to model driver expectations using measured speeds in 153 curves and data on the characteristics of the curve approaches. This model mimics expectations as the probability of measured speeds given certain cues. Using Bayes theorem, prior beliefs on safe speeds are updated towards a posterior belief when a new cue is observed during curve approach. We refer to this posterior belief as expected safe speed. Drivers are assumed to adjust their operating speed if it does not match their expected safe speed. The model shows that the visible deflection angle has a large influence in setting the expectations of a safe speed for an upcoming curve. In addition, the preceding type of roadway and the number of lanes are both important cues to set a driver's expectations of a safe speed. Speed and warning signs are shown to be interdependent on the road scene and hence have less influence in setting expectations. This research shows that design and safety assessment of freeway curves should be considered aligned with the road scene upstream of the curve.

Automated vehicles (AVs) may require the implementation of an external human-machine interface (eHMI) to communicate their intentions to human-driven vehicles. The optimal placement of the eHMI, either on the AV itself or as part of the road infrastructure, remains undetermined. The current driving simulator study investigated the effect of eHMI positioning on human driving behaviour, during the approach and execution of right turns at T-intersections. Forty-three participants drove under three conditions: absence of eHMI, eHMI on the AV (eHMIv), and eHMI integrated into the infrastructure (eHMIi). Participants encountered AVs that either yielded or did not yield to their vehicles. The results regarding the placement of the eHMI showed that both concepts are advantageous, but for different reasons. eHMIv was appreciated because implicit and explicit communication are congruent, although the AV must first be visually identified to respond to it. eHMIi was appreciated because a familiar cue is always at a known location in the environment; as a result, participants braked earlier for the intersection and came less close to the AV (which can be interpreted as a safety advantage or an efficiency disadvantage). Although there are limitations to the current driving simulator study, this research provides important insights into the fundamental question of how information placement affects drivers’ visual attention demands and driving behaviour, topics that are important in view of the development of future cities.

The development and integration of automated driving systems in vehicles hold substantial promise for fostering enhanced efficiency, environmental sustainability, and safety in transportation. Notably, at the lower levels of automation (LI, L2), the lane-keeping system emerges as a widely adopted automated driving feature, ensuring the vehicle’s alignment within its designated lane. With the recent introduction of new European regulations mandating the inclusion of emergency lane-keeping systems in all new vehicles starting July 2022, a growing prevalence of such systems is anticipated in the forthcoming decades. The precision and reliability of these systems in accurately detecting road markings and their distinctive features are paramount for achieving safe and intelligent mobility solutions. To fully capitalize on the advantages these systems offer, they need to expand their operational design domain. This necessitates a comprehensive understanding of their performance across diverse road design and maintenance conditions, supporting road operators in updating standards and maintenance protocols. The primary objective of this study is to investigate how various road characteristics impact the performance of lane-keeping assistant systems. Within this framework, the paper presents an experimental evaluation of Lane-Keeping System (LSS) performance conducted on two-lane rural roads. Advanced technologies for road monitoring and LSS were employed under different road and driving conditions. Through rigorous data analysis and the application of statistical models, variables significant to the fault probability of LSS were identified, highlighting the role played by horizontal curvature and driving speed. Results underscore the relevance of horizontal curvature as a critical factor constraining the physical infrastructure, shaping the operational design domain of LSS. This research contributes valuable insights toward optimizing lane-keeping assistant systems, thereby advancing the development and deployment of safe and efficient automated driving systems in diverse road scenarios.

As automated vehicles (AVs) become increasingly popular, the question arises as to how cyclists will interact with such vehicles. This study investigated (1) whether cyclists spontaneously notice if a vehicle is driverless, (2) how well they perform a driver-detection task when explicitly instructed, and (3) how they carry out these tasks. Using a Wizard-of-Oz method, 37 participants cycled a designated route and encountered an AV multiple times in two experimental sessions. In Session 1, participants cycled the route uninstructed, while in Session 2, they were instructed to verbally report whether they detected the presence or absence of a driver. Additionally, we recorded participants' gaze behaviour with eye-tracking and their responses in post-session interviews. The interviews revealed that 30% of the cyclists spontaneously mentioned the absence of a driver (Session 1), and when instructed (Session 2), they detected the absence and presence of the driver with 93% accuracy. The eye-tracking data showed that cyclists looked more frequently and for longer at the vehicle in Session 2 compared to Session 1. Additionally, participants exhibited intermittent sampling of the vehicle, and they looked at the area in front of the vehicle when it was far away and towards the windshield region when it was closer. The post-session interviews also indicated that participants were curious, but felt safe, and reported a need to receive information about the AV's driving state. In conclusion, cyclists can detect the absence of a driver in the AV, and this detection may influence their perception of safety. Further research is needed to explore these findings in real-world traffic conditions.

This study utilized Virtual Reality (VR) experiments to investigate pedestrian-autonomous vehicle interaction in shared spaces. In the VR experiment, pedestrians attempt to cross the road under different conditions, including the presence of another pedestrian, different external Human-Machine-Interfaces, AV driving styles, and road conditions. We employed an innovative VR setup that enabled two pedestrians to interact in real time with physical movements within an immersive VR environment. Overall, we found that the presence of multiple pedestrians significantly influenced pedestrian movement dynamics during road crossing. Additionally, the relative standing position had a significant impact on the distant pedestrians regarding time before crossing and vehicle-gazing behavior. While previous studies predominantly focused on pedestrian-AV interaction with a single pedestrian, this study takes an important step forward in terms of theory, methods, and relevance by considering interactions between multiple pedestrians and AVs. The findings establish a basis for further exploration of pedestrian-AV interaction in shared space.

Automated Vehicles (AVs) hold promise for revolutionizing transportation by improving road safety, traffic efficiency, and overall mobility. Despite the steady advancement in high-level AVs in recent years, the transition to full automation entails a period of mixed traffic, where AVs of varying automation levels coexist with human-driven vehicles (HDVs). Making AVs socially compliant and understood by human drivers is expected to improve the safety and efficiency of mixed traffic. Thus, ensuring AVs compatibility with HDVs and social acceptance is crucial for their successful and seamless integration into mixed traffic. However, research in this critical area of developing Socially Compliant AVs (SCAVs) remains sparse. This study carries out the first comprehensive scoping review to assess the current state of the art in developing SCAVs, identifying key concepts, methodological approaches, and research gaps. An expert interview was also conducted to identify critical research gaps and expectations towards SCAVs. Based on the scoping review and expert interview input, a conceptual framework is proposed for the development of SCAVs. The conceptual framework is evaluated using an online survey targeting researchers, technicians, policymakers, and other relevant professionals worldwide. The survey results provide valuable validation and insights, affirming the significance of the proposed conceptual framework in tackling the challenges of integrating AVs into mixed-traffic environments. Additionally, future research perspectives and suggestions are discussed, contributing to the research and development agenda of SCAVs.

A shared space area is a low-speed urban area in which pedestrians, cyclists, and vehicles share the road, often relying on informal interaction rules and greatly expanding freedom of movement for pedestrians and cyclists. While shared space has the potential to improve pedestrian priority in urban areas, it presents unique challenges for pedestrian-AV interaction due to the absence of a clear right of way. The current study applied Virtual Reality (VR) experiments to investigate pedestrian-AV interaction in a shared space, with a particular focus on the impact of external human-machine interfaces (eHMIs) on pedestrian crossing behavior. Fifty-three participants took part in the VR experiment and three eHMI conditions were investigated: no eHMI, eHMI with a pedestrian sign on the windshield, and eHMI with a projected zebra crossing on the road. Data collected via VR and questionnaires were used for objective and subjective measures to understand pedestrian-AV interaction. The study revealed that the presence of eHMI had an impact on participants' gazing behavior but not on their crossing decisions. Additionally, participants had a positive user experience with the current VR setting and expressed a high level of trust and perceived safety during their interaction with the AV. These findings highlight the potential of utilizing VR to explore and understand pedestrian-AV interactions.

Detecting abnormal driving behavior is critical for road traffic safety and the evaluation of drivers' behavior. With the advancement of machine learning (ML) algorithms and the accumulation of naturalistic driving data, many ML models have been adopted for abnormal driving behavior detection. Most existing ML-based detectors rely on (fully) supervised ML methods, which require substantial labeled data. However, ground truth labels are not always available in the real world, and labeling large amounts of data is tedious. Thus, there is a need to explore unsupervised or semi-supervised methods to make the anomaly detection process more feasible and efficient. To fill this research gap, this study analyzes large-scale real-world data revealing several abnormal driving behaviors (e.g., sudden acceleration, rapid lane-changing) and develops a Hierarchical Extreme Learning Machines (HELM) based semi-supervised ML method using partly labeled data to accurately detect the identified abnormal driving behaviors. Moreover, previous ML-based approaches predominantly utilize basic vehicle motion features (such as velocity and acceleration) to label and detect abnormal driving behaviors, while this study seeks to introduce Surrogate Safety Measures (SSMs) as the input features for ML models to improve the detection performance. Results from extensive experiments demonstrate the effectiveness of the proposed semi-supervised ML model with the introduced SSMs serving as important features. The proposed semi-supervised ML method outperforms other baseline semi-supervised or unsupervised methods regarding various metrics, e.g., delivering the best accuracy at 99.58% and the best F-1 measure at 0.9913. The ablation study further highlights the significance of SSMs for advancing detection performance.

Autonomous vehicles (AVs) are being introduced to the traffic system with the promise of improving current traffic status. However, the empirical data also indicate contrary effects with estimated higher crash rate and change of crash patterns. Therefore, it is necessary to investigate the driving behavior of AVs and human-driven vehicles (HDVs) in real mixed traffic. Current studies have analyzed the driving behavior of AVs and HDVs, as well as behavioral adaptations of drivers of HDVs based on empirical data. While they play an important role in traffic systems, signalized intersections have not been studied sufficiently in this context. Therefore, this study aims to utilize the Waymo open dataset to characterize and quantify the behavioral differences of AVs and HDVs at signalized intersections. Five parameters of driving behavior related to signalized intersections were characterized according to five critical maneuver phases, which were identified by wavelet transform and threshold-based method. Statistically significant differences in driving behavior between AVs and HDVs were found, from three categorized situations: vehicle approaching the red light/queue, vehicle responding to the green light (as the first vehicle), and vehicle responding to its preceding vehicle (in the queue). Further, behavioral adaptations of HDV drivers were revealed in that they tended to keep closer to the stopped AVs in a queue and to react more strongly to AV start-up maneuvers when the traffic light turns to green.

Traffic scenarios in roundabouts pose substantial complexity for automated driving. Manually mapping all possible scenarios into a state space is labor-intensive and challenging. Deep reinforcement learning (DRL) with its ability to learn from interacting with the environment emerges as a promising solution for training such automated driving models. This study explores, employs, and implements various DRL algorithms, namely Deep Deterministic Policy Gradient (DDPG), Proximal Policy Optimization (PPO), and Trust Region Policy Optimization (TRPO) to instruct automated vehicles' driving through roundabouts. The driving state space, action space, and reward function are designed. The reward function considers safety, efficiency, comfort, and energy consumption to align with real-world requirements. All three tested DRL algorithms succeed in enabling automated vehicles to drive through the roundabout. To holistically evaluate the performance of these algorithms, this study establishes an evaluation methodology considering multiple indicators, i.e., safety, efficiency, comfort and energy consumption level. A method employing the Analytic Hierarchy Process is also developed to weigh these evaluation indicators. Experimental results on various testing scenarios reveal that the TRPO algorithm outperforms DDPG and PPO in terms of safety and efficiency, while PPO performs the best in terms of comfort level and energy consumption. Lastly, to verify the model's adaptability and robustness regarding other driving scenarios, this study also deploys the model trained by TRPO to a range of different testing scenarios, e.g., highway driving and merging. Experimental results demonstrate that the TRPO model trained on only roundabout driving scenarios exhibits a certain degree of proficiency in highway driving and merging scenarios. This study provides a foundation for the application of automated driving with DRL.

This study utilized Virtual Reality (VR) experiments to investigate pedestrian-autonomous vehicle interaction in shared spaces. In the VR experiment, pedestrians attempt to cross the road under different conditions, including the presence of another pedestrian, different external Human-Machin-Interfaces, AV driving styles, and road conditions. We employed an innovative VR setup that enabled two pedestrians to interact in real time with physical movements within an immersive VR environment. Overall, we found that the presence of multiple pedestrians significantly influenced pedestrian movement dynamics during road crossing. Additionally, the relative standing position had a significant impact on the distant pedestrians regarding time before crossing and vehicle-gazing behavior. While previous studies predominantly focused on pedestrian-AV interaction with a single pedestrian, this study takes an important step forward in terms of theory, methods, and relevance by considering interactions between multiple pedestrians and AVs. The findings establish a basis for further exploration of pedestrian-AV interaction in shared space.

Lane detection serves as a fundamental task for automated vehicles and Advanced Driver Assistance Systems. However, current lane detection methods can not deliver the versatility of accurate, robust, and realtime compatible lane detection in real-world scenarios especially under challenging driving scenes. Available vision-based methods in the literature do not consider critical regions of the image and their spatial-temporal salience regarding the detection results, thus they deliver poor performance in peculiar difficult circumstances (e.g., serious occlusion, dazzle lighting). This study aims to introduce a novel sequential neural network model with a spatial-temporal attention mechanism that can focus on key features of lane lines and exploit salient spatial-temporal correlations among continuous image frames for the purpose of enhancing the accuracy and robustness of lane detection. Under the regular encoder-decoder structure and with the implementation using common neural network backbones, the proposed model is trained and evaluated on three large-scale opensource datasets. Extensive experiments demonstrate the strength and the robustness of the proposed model outperforming available state-of-the-art methods in various testing.