Accurately and proactively alerting drivers or automated systems to emerging collisions is crucial for road safety, particularly in highly interactive and complex urban environments. Existing methods require labour-intensive annotation of sparse risk, struggle to consider varying contextual factors or are tailored to limited scenarios. Here we present the generalized surrogate safety measure (GSSM), a data-driven approach that learns collision risk from naturalistic driving without the need for crash or risk labels. Trained on diverse datasets and evaluated on 2,591 real-world crashes and near-crashes, a basic GSSM using only instantaneous motion kinematics achieves an area under the precision–recall curve of 0.9 and secures a median time advance of 2.6 s to prevent potential collisions. Incorporating more interaction patterns and contextual factors provides further performance gains. Across interaction scenarios, such as rear end, merging and turning, GSSM consistently outperforms existing baselines in terms of accuracy and timeliness. These results establish GSSM as a scalable, context-aware and generalizable foundation for identifying risky interactions before they become unavoidable and support proactive safety in autonomous driving systems and traffic incident management.

Max-pressure (MP) control has emerged as a prominent real-time network traffic signal control strategy due to its simplicity, decentralized structure, and theoretical guarantees of network queue stability. Meanwhile, advances in connected vehicle (CV) technology have sparked extensive research into CV-based traffic signal control. Despite these developments, few studies have investigated MP control in heterogeneously distributed and partially CV environments while ensuring network queue stability. To address these research gaps, we propose a CV-based MP control (CV-MP) method that leverages real-time CV travel time information to compute the pressure, thereby incorporating both the spatial distribution and temporal delays of vehicles, unlike existing approaches that utilized only spatial distribution or temporal delays. In particular, we establish sufficient conditions for road network queue stability that are compatible with most existing MP control methods. Moreover, we pioneered the proof of network queue stability even if the vehicles are only partially connected and heterogeneously distributed, and gave a necessary condition of CV observation for maintaining the stability. Evaluation results on an Amsterdam corridor show that CV-MP significantly reduces vehicle delays compared to both actuated control and conventional MP control across various CV penetration rates. Moreover, in scenarios with dynamic traffic demand, CV-MP achieves lower spillover peaks even with low and heterogeneous CV penetration rates, further highlighting its effectiveness and robustness.

Max-pressure (MP) control has proven effective at stabilizing network queues and improving traffic throughput in large-scale urban road networks. However, conventional MP controllers based on connected vehicle (CV) data face two critical limitations: network stability diminishes when connected vehicle (CV) penetration rates are low, and significant privacy concerns arise when utilizing individual vehicle data. To address these challenges, this paper proposes a novel Private-MP controller that fuses data from both fixed-location detectors and CVs in an architecture of mobile edge computing. To fully safeguard CV privacy, including macro-route information and micro-trajectory information, Private-MP employs a privacy-preserving mechanism that combines homomorphic encryption with an adaptive randomized response strategy. Simulation studies on a network with five intersections showed that despite some increases in average vehicle delay due to privacy protection, Private-MP still ensures a more robust performance on average vehicle delay than CV-based MP in low penetration rate scenarios and outperforms traditional detector-based MP control while improving fairness among connected and non-connected vehicles.

Large-scale prediction of trip production is essential for origin–destination (OD) demand estimation and prediction. One of the main challenges in predicting trip production patterns lies in addressing spatial-temporal correlations and variations. Whereas many studies focus on temporal correlations, very few consider spatial adjacency between traffic analysis zones (TAZ) as explanatory variables. This research proposes a method that integrates a graph convolutional neural network (GCN) into a long short-term memory network (LSTM) to do exactly that. By introducing a nationwide graph that encodes the adjacency of TAZs, spatial heterogeneity is considered in the prediction process, and a single prediction model is trained for the entire network, thereby avoiding the need to train multiple separate models and potentially reducing overall training overhead, while increasing the prediction accuracy. Moreover, with this model, we investigate the effect of spatial scale on spatial uncertainty and prediction accuracy and analyze prediction errors, residual patterns, and their associations with socio-spatial features at different spatial scales. The findings of this research have important implications for improving OD demand prediction models and provide valuable insights into the role of spatial scale and socio-spatial features in travel demand prediction.

The operation of intelligent connected vehicles (ICVs) is fundamentally data-driven, continuously generating massive amounts of data. Given the significant value of ICV data to enterprises, industries, and nations, promoting data openness and sharing has become essential. However, such data often contain sensitive information, and its misuse can threaten individual privacy, corporate security, and even national interests. To address this dilemma, this paper develops the misuse risk score (MR-score), a novel quantification model and associated evaluation method for assessing the risk of ICV data misuse. The MR-score is constructed based on three core properties of ICV data: sensitivity; scale; and identifiability. The sensitivity score, information quantity, and identifiability factor are designated as the corresponding evaluation indicators, and systematic approaches for their quantification are proposed. The analytic hierarchy process is employed to measure the sensitivity score. Information entropy is adopted to evaluate the information quantity. A combination of k-anonymity-based and damage source determination-based methods is utilized to estimate the identifiability factor, considering data incompleteness, imprecision, and invalidity. Two empirical ICV data sets are utilized, and comparative analyses are conducted to demonstrate the effectiveness of the MR-score in capturing misuse risks. Higher MR-scores correspond to greater risk. The model captures the joint influence of all three data properties and reveals the marginal diminishing effect of data scale on misuse risk. This work offers valuable tools for data owners and regulatory agencies to prioritize critical data sets, implement targeted data protection measures, and enable secure data circulation while maximizing the value of ICV data.

Conditionally automated driving requires drivers to resume vehicle control within constrained time budgets upon receiving takeover requests. Accurately predicting drivers’ takeover time (ToT) is essential for dynamically adjusting time budgets to individual needs across scenarios. This study addresses enduring challenges in reliability and interpretability of ToT prediction models by optimizing predictor selection. Using a driving simulator experiment, we examine the relationship between ToT, driver characteristics, and perceived Spare Capacity (pSC, a cognitive construct from Task-Capability Interface theory) using Category Boosting models. Results show that (i) incorporating 13 additional driver characteristics does not significantly improve prediction accuracy when pSC is already considered; and (ii) individual characteristics influence how drivers cognitively process takeover scenarios, and their predictive contribution likely overlaps with pSC. These findings suggest that monitoring cognitive states may be more effective for ToT prediction than extensive profiling of driver characteristics. This study provides a critical first step toward predictive frameworks for adaptive takeover strategies and offers guidance for designing personalized human–vehicle interactions.

Integrating renewable energy sources, such as solar and wind, challenges grid stability due to their intermittent nature. Vehicle-to-grid (V2G) technology provides a promising solution by utilizing electric vehicles (EVs) as decentralized energy storage systems, enabling the storage of surplus energy during low demand and its release during peak demand. The effectiveness of V2G depends critically on car usage patterns. Data from the Netherlands Mobility Panel (MPN) of 2022, comprising travel diaries from 2,505 households, was analyzed to explore this. A methodology was developed to create car usage profiles based on parking durations and locations, distinguishing weekday and weekend patterns. The analysis shows that vehicles are predominantly parked at home, with weekday profiles reflecting work-related parking and weekend profiles highlighting increased leisure activity. Households with shared cars showed higher driving activity and shorter parking durations than households with a 1:1 car-to-license ratio or surplus vehicles. Six distinct car usage clusters were identified for weekdays and four for weekends.

Smart traffic systems, like those using wellestablished methods such as SCOOT, SCATS and TUC, aim to improve traffic flow by dynamically adjusting signal timings based on real-time traffic conditions. Traffic engineers need to understand the objective functions behind traffic signal control to analyze, improve, and optimize network performances. However, different jurisdictions, different operators and competing interests imply that the underlying objective functions governing traffic signal control might not be publicly known with sufficient detail (e.g. to preserve Intellectual Property Rights). A method for discovering these functions is therefore needed, particularly to enable better cooperation among stakeholders. In this work, we train computer models to mimic the decisions made by smart traffic light systems. Using data from a simulated traffic network (with virtual sensors tracking vehicles), we test a variety of supervised models, ranging from simple decision trees to more complex neural networks. Our results show these models can accurately mimic the underlying system's actions, achieving up to 99% accuracy. This work demonstrates that supervised learning can serve as a powerful tool for uncovering hidden traffic control functions by training models to replicate the system's decisions. By analyzing these models, we can then infer the key factors influencing signal control, thereby gaining insights into the underlying objective function.

Predicting the trajectories of road agents is fundamental for self-driving cars. Trajectory prediction contains many sources of uncertainty in data and modelling. A thorough understanding of this uncertainty is crucial in a safety-critical task like auto-piloting a vehicle. In practice, it is necessary to distinguish between the uncertainty caused by partial observability of all factors that may affect a driver's near-future decisions, the so-called aleatoric uncertainty, and the uncertainty of deploying a model in new scenarios that are possibly not present in the training set, the so-called epistemic uncertainty. They reflect the trade-off between data collection and model improvement In this paper, we propose a new framework to systematically quantify both sources of uncertainty. Specifically, to approximate the spatial distribution of an agent's future position, we propose a 2D histogram-based deep learning model combined with deep ensemble techniques for measuring aleatoric and epistemic uncertainty by entropy-based quantities. The proposed Uncertainty Quantification Network (UQnet) employs a causal part to enhance its generalizability so rare driving behaviours can be effectively identified. Experiments on the INTERACTION dataset show that UQnet is able to give more robust predictions in generalizability tests compared to the correlation-based models. Further analysis presents that high aleatoric uncertainty cases are mainly caused by heterogeneous driving behaviours and unknown intended directions. Based on this aleatoric uncertainty component, we estimate the lower bounds of mean-square-error and final-displacement-error as indicators for the predictability of trajectories. Furthermore, the analysis of epistemic uncertainty illustrates that domain knowledge of speed-dependent driving behaviour is essential for adapting a model from low-speed to high-speed situations. Our paper contributes to motion forecasting with a new framework, that recasts the problem of accuracy improvement in a way that focuses on differentiating between unpredictable components and rare cases for which more and different data should be collected.

Safety is the cornerstone of L2+ autonomous driving and one of the fundamental tasks is forward collision warning that detects potential rear-end collisions. Potential collisions are also known as conflicts, which have long been indicated using Time-to-Collision with a critical threshold to distinguish safe and unsafe situations. Such indication, however, focuses on a single scenario and cannot cope with dynamic traffic environments. For example, TTC-based crash warning frequently misses potential collisions in congested traffic, and issues false alarms during lane-changing or parking. Aiming to minimise missed and false alarms in conflict detection, this study proposes a more reliable approach based on vehicle spacing patterns. To test this approach, we use both synthetic and real-world conflict data. Our experiments show that the proposed approach outperforms single-threshold TTC unless conflicts happened in the exact way that TTC is defined, which is rarely true. When conflicts are heterogeneous and when the information of conflict situation is incompletely known, as is the case with real-world conflicts, our approach can achieve less missed and false detection. This study offers a new perspective for conflict detection, and also a general framework allowing for further elaboration to minimise missed and false alarms. Less missed alarms will contribute to fewer accidents, meanwhile, fewer false alarms will promote people's trust in collision avoidance systems. We thus expect this study to contribute to safer and more trustworthy autonomous driving.

Resolving predicted conflicts is vital for safe and efficient autonomous vehicles (AV). In practice, vehicular motion prediction faces inherent uncertainty due to heterogeneous driving behaviours and environments. This spatial uncertainty increases non-linearly with prediction time horizons, leading AVs to perceive more road space occupied by conflicting vehicles. Reacting early to resolve predicted conflicts can ensure safety but may adversely affect traffic efficiency. Therefore, determining how far ahead AVs should start resolving predicted conflicts based on safety and traffic efficiency constraints is crucial. To answer this question, this study proposes a novel approach to explore the trade-off between safety and traffic efficiency considering prediction uncertainty. Firstly, a continuous-time motion prediction framework is proposed for estimating the spatial probability distribution of a vehicle's future position at any moment within the maximum time horizon. Subsequently, average driver space and the corresponding traffic flow are derived from the safety settings of AV and prediction uncertainty. As such, the safety-efficiency trade-off can be quantified. Experiments show that mandatory decision points, high speeds, and traffic state transitions usually cause fast-increasing prediction uncertainty. A case study of Intelligent Driver Models (IDM) shows that traffic efficiency drops rapidly when AVs resolve predicted conflicts longer than 1.5 seconds ahead. AVs can act earlier on motorways for efficiency concerns but must be myopic at urban intersections. Prediction uncertainty fundamentally constrains the safety-efficiency performance of AVs. These findings are instructive for designing traffic-compatible AVs.

Traffic conflict detection is essential for proactive road safety by identifying potential collisions before they occur. Existing methods rely on surrogate safety measures tailored to specific interactions (e.g., car-following, side-swiping, or path-crossing) and require varying thresholds in different traffic conditions. This variation leads to inconsistencies and limited adaptability of conflict detection in evolving traffic environments, particularly as the integration of autonomous driving systems adds complexity. Consequently, there is an increasing need for consistent detection of traffic conflicts across interaction contexts. To address this need, we propose a unified probabilistic approach in this study. The proposed approach establishes a unified framework of traffic conflict detection, where traffic conflicts are formulated as context-dependent extreme events of road user interactions. The detection of conflicts is then decomposed into a series of statistical learning tasks: representing interaction contexts, inferring proximity distributions, and assessing extreme collision risk. The unified formulation accommodates diverse hypotheses of traffic conflicts and the learning tasks enable data-driven analysis of factors such as motion states of road users, environment conditions, and participant characteristics. Jointly, this approach supports consistent and comprehensive evaluation of the collision risk emerging in road user interactions. We demonstrate the proposed approach by experiments using real-world trajectory data. A unified metric for indicating conflicts is first trained with lane-change interactions on German highways, and then compared with existing metrics using near-crash events from the U.S. 100-Car Naturalistic Driving Study. Our results show that the unified metric provides effective collision warnings, generalises across distinct datasets and traffic environments, covers a broad range of conflict types, and captures a long-tailed distribution of conflict intensity. In summary, this study provides an explainable and generalisable approach that enables traffic conflict detection across varying interaction contexts. The findings highlight its potential to enhance the safety assessment of traffic infrastructures and policies, improve collision warning systems for autonomous driving, and deepen the understanding of road user behaviour in safety–critical interactions.

Limited available market share data seems to suggest that ridesourcing platforms benefit from, even thrive on, socio-economic inequality. We suspect that this is associated with high levels of socio-economic inequality allowing for cheap labour as well as increasing the share of travellers with a considerably above-average willingness to pay for travel time savings and comfort. We test the relation between inequality and system performance by means of an agent-based simulation model representing within-day and day-to-day supply-demand interaction in the ridesourcing market. The model captures travellers’ mode choice with a heterogeneous perception of relevant time components, as well as job seekers’ participation choice with heterogeneous reservation wage. Our experiments cover scenarios for the entire spectrum ranging from perfect equality to extreme inequality. For several of such scenarios, we explore alternative platform pricing strategies. Our analysis shows a strong, positive relationship between socio-economic inequality and ridesourcing market share. This is the outcome of the combination of cheap labour and time-sensitive ridesourcing users, reinforced by network effects inherent to ridesourcing markets. We find that driver earnings are minimal in urban areas with large socio-economic inequality. In such contexts, drivers are likely to face a high platform commission, and yet, fierce competition for passengers.

Integrating Automated Vehicles (AVs) into existing traffic systems holds the promise of enhanced road safety, reduced congestion, and more sustainable travel. Effective integration of AVs requires understanding the interactions between AVs and Human-driving Vehicles (HVs), especially during the transition period in which AVs and HVs coexist in a mixed traffic environment. Numerous recent empirical studies find reduced headways of human drivers following an AV compared to following an HV, and attribute this reduction to behavioural changes of drivers when they follow AVs. However, more factors may be at play due to the inherent inconsistencies between the comparison conditions of HV-following-AV and HV-following-HV. This study scrutinises three alternative explanations for the observed reduction in headways: (1) systematic differences in car-following states during data collection, (2) systematic differences in driving variability between leading AVs and HVs, and (3) systematic differences in driving characteristics of leading AVs versus HVs. We use a large-scale dataset extracted from Lyft AV motion data and examine each of these explanations through data stratification and simulation. Our results show that all three mechanisms contribute to the observed reduction in headways of human drivers following AVs. In addition, our findings highlight the importance of driving homogeneity and stability in achieving reliably shorter headways. Thereby, this study offers a more comprehensive understanding on the difference between HV–AV and HV–HV interactions in mixed traffic, and is expected to promote more effective integration of AVs into human traffic.