Autonomous vessels are becoming paramount to ocean transportation, while they also face complex risks in dynamic marine environments. Machine learning plays a crucial role in enhancing maritime safety by leveraging its data analysis and predictive capabilities. However, there has been no review grounded in bibliometric analysis in this field. To explore the research evolution and knowledge frontier in the field of maritime safety for autonomous shipping, a bibliometric analysis was conducted using 719 publications from the Web of Science database, covering the period from 2000 up to May 2024. This study utilized VOSviewer, alongside traditional literature analysis methods, to construct a knowledge network map and perform cluster analysis, thereby identifying research hotspots, evolution trends, and emerging knowledge frontiers. The findings reveal a robust cooperative network among journals, researchers, research institutions, and countries or regions, underscoring the interdisciplinary nature of this research domain. Through the review, we found that maritime safety machine learning methods are evolving toward a systematic and comprehensive direction, and the integration with AI and human interaction may be the next bellwether. Future research will concentrate on three main areas: evolving safety objectives towards proactive management and autonomous coordination, developing advanced safety technologies, such as bio-inspired sensors, quantum machine learning, and self-healing systems, and enhancing decision-making with machine learning algorithms such as generative adversarial networks (GANs), hierarchical reinforcement learning (HRL), and federated learning. By visualizing collaborative networks, analyzing evolutionary trends, and identifying research hotspots, this study lays a groundwork for pioneering advancements and sets a visionary angle for the future of safety in autonomous shipping. Moreover, it also facilitates partnerships between industry and academia, making for concerted efforts in the domain of USVs.

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.

Collision avoidance in maritime navigation, particularly between autonomous and conventional vessels, involves iterative and dynamic processes. Traditional path planning models often neglect the behaviours of surrounding vessels, while path predictive models tend to ignore ship interaction features essential in collision scenarios. This study proposes a decision-making framework for collision avoidance, particularly focused on the interaction between autonomous and conventional vessels. The framework integrates a human-preference-aware navigational trajectory prediction model into the collision avoidance process, enhancing the decision-making capabilities in dynamic and interactive environments. We first model human-controlled ship navigational preferences using a Long Short-Term Memory (LSTM) autoencoder combined with K-means clustering, by extracting key preferences from ship pairs identified through AIS data. These preferences, which reflect strategic trajectory adjustments in response to collision risks, are then incorporated into trajectory prediction using a Multi-Task Learning Sequence-to-Sequence (seq2seq) attention LSTM model. The predicted trajectories provide a basis for the decision-making framework, including a local path planner and a trajectory tracking controller, designed to dynamically adjust the predicted reference path for collision-free navigation and ensure its accurate tracking. The framework was validated using AIS data from the port of Rotterdam, identifying four distinct navigational preferences by combining an LSTM-autoencoder and clustering techniques and demonstrating improved prediction accuracy compared to other existing models. Simulation tests demonstrate that the framework utilises the predicted trajectories to inform decision-making, ensuring accurate path tracking while dynamically addressing collision risks for autonomous ships. By providing preference-aware and adaptive reference trajectories, the framework reduces the likelihood of MASS trajectory misinterpretation by conventional ships, thereby supporting proactive collision avoidance in mixed waterborne transport environments.

Human–Machine Interfaces (HMIs) in passenger cars have become more complex over the years, with touch screens replacing physical buttons and with layered menu-structures. This can lead to distractions. The purpose of this study is to investigate how often vehicle controls are used while driving and which underlying factors contribute to usage. Thirty drivers were observed during driving a familiar route twice, in their own car and in an unfamiliar car. In a 2 × 1 within-subject design, the experimenter drove along with each participant and used a predefined checklist to record how often participants interacted with specific functions of their vehicle while driving. The results showed that, in the familiar car, direction indicators are the most frequently used controls, followed by adjusting radio volume, moving the sun visor, adjusting temperature and changing wiper speed. Factors that influenced task frequencies included car familiarity, gender, age and weather conditions. The type of car also appears to impact task frequency. Participants interacted less with the unfamiliar car, compared to their own car, which may indicate drivers are regulating their mental load. These results are relevant for vehicle HMI designers to understand which functions should be easily and swiftly available while driving to reduce distraction by the HMI design.

Automated Driving Systems (ADS) are aimed to improve traffic efficiency and safety, however these systems are not yet capable of handling all driving tasks in all types of road conditions. The role of a human driver remains crucial in taking over control, if an ADS fails or reaches its operational limits. Takeover performance of human drivers in authority transitions is typically assessed by means of the takeover time (TOT) taken within an available time budget (TB). This approach assumes a uniform perception and reaction time of human drivers in ADS disengagements, and does not include the time needed to execute the actual driving maneuver required to ensure safety. This paper aims to develop and test a set of new indicators to reflect takeover performance and its safety attributes, namely the ‘time to control’ (TC) and the ‘safe time budget’ (STB), in which the actual task execution (i.e. braking) time is taken into account, in addition to the perception and reaction time. It also proposes new thresholds for identifying critical conflicts in takeover situations and assessing the safety of authority transitions. A traffic simulation experimental setup is used with mixed traffic of conventional vehicles and ACC/CACC platoons in order to test these indicators and thresholds. The results suggest that the time difference between TC and STB is a more sensitive and potentially more realistic safety indicator, as it may capture the variability of driver behavior in takeovers and identify critical conflicts, as well as virtual crashes, that would not have been identified by the previously used indicators (TOT and TB). Takeover performance worsens when the speed difference of the vehicles involved is higher, and the initial speed of the rear vehicle is higher. These findings can be useful towards a more dynamic design of takeover request strategies.

Maritime Autonomous Surface Ships (MASS) have gained much attention as a safer and more efficient mode of transportation and a potential solution to reduce the workload of seafarers. Despite the highly sophisticated autonomous systems that enable MASS to make independent decisions, the presence of humans on board or in the loop of safety management highlights the need for effective human-machine interaction. However, a potentially systematic review of critical aspects of human-MASS interaction has not yet been conducted. In this paper, we aim to fill this gap by reviewing the literature related to human-MASS interaction from four crucial perspectives: the state of the art of human-MASS interaction, situational awareness for MASS, collision avoidance methods for MASS within a mixed waterborne transport system (MWTS), and human trust in MASS. Our review reveals that human-MASS interaction for safety and efficiency mainly focuses on four key aspects: (i) human factors, (ii) available technologies supporting the autonomy of MASS, (iii) system analysis and design for human-MASS interaction, and (iv) potential requirements regarding regulations. Moreover, we provide a detailed discussion of the three fundamental factors that influence human-MASS interaction, including situational awareness, decision-making for MASS in a mixed waterborne transport system, and human trust in the autonomous system of MASS. Finally, based on our analysis, we propose an integrated framework of human-MASS interaction in which these human factors are taken into account. We anticipate that these factors and their interaction will receive more attention to improve the safety and efficiency of MASS.

Existing models for correlating global mortality rates with underlying country-specific factors overlook the variations in the effects of these factors on mortality across different countries. These may arise from social, cultural, and political complexities which are usually not measurable and are therefore referred to as unobserved heterogeneity in the statistical literature. Unobserved heterogeneity leads to biased parameter estimates in the models, erroneous inferences about the effects of factors contributing to mortalities, and ultimately inefficient policies. In this paper, latent class modelling is proposed for capturing such unobserved heterogeneity on the cases of traffic mortality and COVID-19 mortality. The ‘pyramid’ model of safety management is used as a common framework for model formulation. The proposed latent class model is an extension of the Negative Binomial (NB) model used in risk epidemiology. The model is tested with data from 105 countries, retrieved from international databases, including socioeconomic, infrastructure, exposure, transport, and COVID-19 variables. The results suggest that there exist two (different) latent country classes in both causes of mortality. The probability of a country belonging to a certain latent class is a much more efficient metric of country membership than previous deterministic groupings (e.g. income or geographic). Variables such as the elderly population, the GDP per capita or the level of motorization, have different effects in different country classes; these effects are not identifiable by conventional statistical modelling. The impact of ignoring unobserved heterogeneity in country mortality modelling is shown by comparing the results with those of conventional NB models.

This study investigates the enhancement of Maritime Autonomous Surface Ships (MASS) navigation and path-planning through the integration of ontology-based knowledge maps (KM) with the Dynamic Window Approach (DWA), a fusion termed KM-DWA. The ontology-based KM model is important for MASS navigation, offering a framework for situational awareness, including contextual information fusion and decision-making evidence. This research enriches the KM model with collision avoidance rules from the International Regulations for Preventing Collisions at Sea (COLREGs), building upon our previous work on MASS's efficient and COLREGs-compliant navigation in encounter scenarios. The model provides navigational context, covers COLREGs rules and environmental factors, and recommends MASS actions for various scenarios as suggested by COLREGs. Moreover, an adapted DWA, tailored to maritime navigation, accounts for specific constraints and safety measures for MASS, utilising KM-derived situational awareness as constraints in its cost function for path planning. A significant innovation introduced here is a tiered safety distance model featuring proactive, defensive, and collision buffers to ensure rule-compliant and effective collision avoidance. This scheme enables MASS to take timely collision avoidance actions at both proactive and defensive distances, in line with COLREGs recommendations. The effectiveness of the KM-DWA algorithm is validated by comparing it with the basic DWA algorithm in single- and multi-vessel encounter scenarios. The experiment outcomes illustrate the integrated approach's superiority in terms of COLREGs compliance and collision avoidance rate, emphasising its ability to support COLREGs-compliant decision-making and enhance situational awareness in autonomous maritime operations.

The i-DREAMS project focuses on establishing a framework known as the ‘Safety Tolerance Zone (STZ)’ to ensure drivers operate within safe boundaries. This study compares Long-Short-Term-Memory Networks and shallow Neural Networks to assess participants’ safety levels during i-DREAMS on-road trials. Thirty German drivers’ trips and Forty-Three Belgian drivers were analyzed using these methods, revealing factors contributing to risky behavior. Results indicate i-DREAMS interventions significantly enhance driving behavior, with Neural Networks displaying superior performance among the algorithms considered.

Naturalistic driving studies (NDS) have recently gained attention as a way of instrumenting vehicles in an unobtrusive way and collecting driving data over long periods of time. Aiming at eventually modeling driving behavior, NDS are often a part of larger scale studies. These studies involve several stakeholders who are responsible for different components of the data collection and analysis, and thus are inevitably confronted with challenges in the data management pipeline. The aim of this paper is to develop standard protocols that could be used as guidelines for data handling in the context of NDS. In the development of these protocols, we first review data handling strategies used in previous studies, focusing on data collection, preparation, storage, as well as ethical and legal considerations. This review helps us draw lessons, based on which methods are developed to answer the gaps and challenges arising from handling NDS data. We then introduce a case study, the i–DREAMS project, to show the applicability of the data handling framework. Finally, we showcase standard protocols for data handling, that could serve as data handling guidelines for future studies.

Although signal countdown timers (SCTs) are likely to enhance efficiency at signalized intersections, there is little research on how they affect road users’ behavior. The present study explores factors associated with driver behavior through two approaches to examine how SCTs influence drivers’ actions toward pedestrians violating red lights. In the first approach, through an on-road questionnaire survey, the self-reported behavior of 369 drivers when crossing an intersection enabled with SCTs was analyzed. In the second approach, the drivers’ behavior was studied through naturalistic driving studies at two signalized intersections equipped with SCTs in Babol, Iran. Analyzing vehicle–pedestrian conflicts indicated that the presence of SCTs had a significant influence on driving behavior. Also, the ending seconds of green lights, as critical times of the SCTs, led to changes in driving behavior. Increasing the vehicle speed, changing lanes, and concurrent increases of speed and changing lanes were the common driver actions affected by critical times of the SCTs. Finally, the effect of critical times on drivers’ actions during conflicts was modeled by using the binary and multinomial logistic methods. The results show that SCTs are an external factor that can lead to risky driver behavior, such as errors and violations that might increase the potential for pedestrian accidents.

Recent research in transport safety focuses on the processing of large amounts of available data by means of intelligent systems, in order to decrease the number of accidents for transportation users. Several Machine Learning (ML) and Artificial Intelligence (AI) applications have been developed to address safety problems and improve efficiency of transportation systems. However exchange of knowledge between transport modes has been limited. This paper reviews the ML and AI methods used in different transport modes (road, rail, maritime and aviation) to address safety problems, in order to identify good practices and experiences that can be transferable between transport modes. The methods examined include statistical and econometric methods, algorithmic approaches, classification and clustering methods, artificial neural networks (ANN) as well as optimization and dimension reduction techniques. Our research reveals the increasing interest of transportation researchers and practitioners in AI applications for crash prediction, incident/failure detection, pattern identification, driver/operator or route assistance, as well as optimization problems. The most popular and efficient methods used in all transport modes are ANN, SVM, Hidden Markov Models and Bayesian models. The type of the analytical technique is mainly driven by the purpose of the safety analysis performed. Finally, a wider variety of AI and ML methodologies is observed in road transport mode, which also appears to concentrate a higher, and constantly increasing, number of studies compared to the other modes.

This study reviews the Artificial Intelligence and Machine Learning approaches developed thus far for driver profile and driving pattern recognition, representing a set of macroscopic and microscopic behaviors respectively, to enhance the understanding of human factors in road safety, and therefore reduce the number of crashes. It provides a definition of the two scientific fields in terms of safety, and identifies the most efficient approaches used regarding methodology, data collection and driving metrics. Results show that K-means and Neural Networks are the most commonly used methodologies for driver profile identification, and Dynamic Time Warping for driving pattern detection. Most studies discovered driver profiles related to aggressiveness, considering mainly speed and acceleration as driving metrics. Based on the gaps and challenges identified, this paper provides a new framework for combining microscopic and macroscopic driving behavior analysis, instead of examining them separately as is the state-of-theart. Such combined results can potentially improve the development of traffic risk models, which could be exploited in applications that monitor drivers in real-time and provide feedback. These models will represent human behavior more accurately, which can eventually lead to the recognition of 'optimal' human driving patterns that Automated Vehicles (AV) could 'mimic' to become safer.

The safety of maritime autonomous surface ships (MASS) in mixed waterborne transport system (MWTS) depends on effective situational awareness (SA) distribution among MASS, manned ships, and various stakeholders, such as Vessel Traffic Service (VTS), Remote Control Center (RCC) and Fairway Shipping Agency. This paper focuses on the research question: How can situational awareness be effectively distributed among these entities in mixed waterborne transport? The research objective is to develop a distributed situational awareness framework that unifies SA among these stakeholders, ensuring safe navigation and compatibility with users of different roles. To achieve this objective, the proposed framework incorporates three key concepts: individual SA, authority-based SA, and distributed SA. Individual SA, previously introduced in our study, is responsible for each ship's SA, while authority-based SA accounts for the SA of human operators supervising the waterborne transport system, such as VTS operators and fairway agency personnel. Distributed SA generates guiding messages for ships based on the situational awareness from both individual SA and authority-based SA, thereby enabling regulation-based and traffic control-based recommendations for waterborne transport (e.g., ship speed and course adjustments). The research methodology employs ontology-based modelling to implement the framework, constructing a domain knowledge network. A case study is conducted as an essential part of the research methodology, presenting how the framework perform the situational awareness from different aspects and inconsistency detection among manned ships, MASS, VTS operators, and so on. Semantic Web Rule Language (SWRL) is utilized to detect inconsistencies and generate guidance messages for ships. Through these cases, we demonstrate how the proposed Ontology-based framework can reconcile inconsistencies between individual and authority-based SA, leading to a safer and more effective waterborne transport.

Driving pattern recognition has been applied for the purposes of driving styles identification and harsh driving events detection. However, the evolution of driving behavior around and especially before such events has not been investigated at a microscopic level. The objective of this research is to reveal existing driving patterns around harsh events at the driving ‘pulse’ level i.e. a few seconds before and after the event. For that purpose, a time-series clustering approach is applied on speed and heart rate metrics of individual drivers using data collected from a large naturalistic driving study. Results show that there are distinct speed patterns before harsh braking, harsh acceleration, and harsh cornering events. A deceleration is identified shortly before most harsh acceleration and cornering events, which possibly indicates reckless behavior, i.e. drivers not dedicating enough time to smoothly brake before cornering, or of a brief ‘decision-making’ moment before the harsh manoeuvre. On the contrary, speed seems to be steady before harsh braking events. Regarding heart rate, the analysis revealed certain patterns only after raw data were cleansed and filtered. These patterns may show increasing, decreasing or variable heart rate trends, which may correspond to different stress patterns of drivers around harsh events. Finally, we introduce the concept of driving pattern consistency, which can reveal the share of individual drivers that follow the same harsh event pattern. It is indicated that more than half of the drivers are not consistent, suggesting that driving patterns around harsh events may be more context-related than driver personality-related.

Driver anticipation plays a crucial role in crashes along horizontal curves. Anticipation is related to road predictability and can be influenced by roadway geometric design. Therefore, it is essential to understand which geometric design elements can influence anticipation and cause the road to be (un)predictable. This exercise, however, is not straightforward because anticipation is individual-specific whereas road geometric design is location-specific; anticipation is latent and measuring it may not be trivial; anticipation may have several stages from the preceding tangent until the midst of the curve; and not all drivers anticipate in the same way and thus there may well be unobserved heterogeneity in the effect of anticipation on crash risk. Despite methodological advancements in crash risk modelling, there is no econometric model that can adequately explain the above complexities. This study aims to fill this gap by developing an econometric model with a new latent variable, named ‘predictability’ that is measured by individual-specific driving behaviour indicators and predicted by location-specific road geometric factors. The model is specified with random parameters to account for unobserved heterogeneity and is empirically tested by a unique dataset including detailed geometric design and driver behaviour data obtained for 156 curves in the Netherlands. Results indicate that higher exposure and uphill vertical grade are associated with increased likelihood of vehicle crashes along horizontal curves, whereas adequate superelevation and higher predictability are associated with decreased likelihood of those crashes. Pavement friction influences this likelihood too but it has varied effects. Road predictability is influenced by the differences in angle of horizontal curves, vertical grades, and width of consecutive road segments.

Driver behavior analytics is an important concept that plays a significant role in the understanding of road crashes. This paper investigates the optimal number of driver profiles to understand the most important characteristics that differentiate drivers and extract useful insights on the value of using different clustering approaches in profile recognition. To this end, two Machine Learning clustering algorithms, the K-Means and OPTICS algorithms, are applied on driving data from a large naturalistic experiment using almost 18 K trips recorded from 130 drivers. The results revealed 3 profiles, the less risky drivers, the modest drivers and the more aggressive drivers. Clustering was based on 3 important driving behavior characteristics, namely the number of speeding, headway and harsh events per 100 km. The less risky drivers profile was revealed by both algorithms, whereas drivers of higher aggressiveness are distinguished by K-Means based on the driving feature that dominates the rest. The OPTICS algorithm showed that many drivers, especially the aggressive ones, present unique behavior that cannot be grouped together with other drivers. The interpretability of driver profiles resulting from the application of these unsupervised learning techniques is worsened as the number of clusters increases. The association between driver profiles and individual characteristics leads to the conclusion that aggressiveness is mainly driven by personality traits and less by specific characteristics such as gender, age or past accident history. The results of this study can be potentially used to develop profile-specific applications that provide feedback to drivers and reduce their crash risk.

Many projects related to maritime autonomous surface ships (MASS) have been proceeding to date, which promotes the commercialization of MASS. It is anticipated that there will be ships with different degrees of autonomy coexisting in a waterborne transport system (WTS) in the near future, forming a mixed waterborne transport system (MWTS). To ensure navigational safety, the ship needs to be well aware of the situation in real time, i.e. be able to exhibit situation awareness. The inconsistency of comprehension within SA of the same situation could occur in the MWTS, leading to a potentially dangerous situation. As such, it is essential to unify the SA framework for MASS to eliminate the inconsistency with human operators. It is challenging but necessary for a MASS to accomplish the process of situation awareness involving perception, comprehension, and projection. Especially the part of comprehension is the core element that needs to be addressed well and enhanced further. One possible way to reach it is to integrate the information given by the perception layer, projection layer, as well as additional domain knowledge like navigational rules to conduct further analysis. Accordingly, the current paper proposes a method for knowledge integration of SA for the MASS. The method realizes four capabilities of SA to satisfy relevant requirements in maritime domain: a general map, risk assessment enrichment, temporal and dynamic features, as well as a supplement of domain knowledge. For that purpose, the paper takes two steps: (i) constructing a SA framework for MASS, in which the entities related to SA are classified to different categories. (ii) proposing an ontology-based SA comprehension model where the information of entities are integrated together and then the SA can be depicted for MASS in real time. A case study is provided to how the model can be applied in maritime domain, in which a MASS is approaching a port executing its tasks. As a result, the proposed method can relate the information provided by both the perception and projection layers, and domain knowledge in the form of a knowledge graph to depict the real-time situation. The results show that the method is feasible to provide potentials to the MASS to be aware of the situation in real time considering domain knowledge. The method can be applied to MASS for the information fusion of situation awareness, which also can be supplied to Maritime Safety and Security (MSS) organizations for traffic surveillance of WTS.