ABSTRACT. Modeling dynamic urban traffic accurately requires travel demand data at high temporal and spatial resolution. Congestion results from overlapping and time-dependent trips that cannot be represented adequately by static demand formats such as temporally aggregated OD matrices. At the same time, many national forecasting tools estimate demand using aggregated data and simplified behavioral representations. We propose a kernel-based framework for incorporating individual-level travel survey information into the simulation of agent travel plans. Kernel similarity measures combine socio-demographic characteristics, trip attributes, and activity--location information, making it possible to compare heterogeneous objects such as travelers, trips, and zones across different feature spaces. Such an approach is interpretable and modular, making it attractive for travel-demand synthesis, where comparisons across diverse domains are required.

We demonstrate the approach in a small case study for Stockholm, using the national travel survey and a synthetic population applied in real-world transport policy analysis. The results indicate that key marginal characteristics of the survey data are reproduced in the simulated trips. Departure-time distributions, which are crucial for urban congestion analysis, were represented well despite not being explicitly encoded in the model. The activity--location component further combines survey-based activity participation with external land-use information to guide where activities are undertaken.

ABSTRACT. This paper represents a first step towards a graph-based, econometric activity modelling framework, incorporating both location and spatial choices into the behavioural process. Current econometric activity based models typically focus predominantly on the temporal aspects of activity scheduling, leaving spatial decisions to agent-based microsimulation frameworks. Whilst recent work has established behavioural frameworks that incorporate location choice into the scheduling process, they have not yet been applied practically as (i) joint schedule optimisation across large numbers of candidate locations is computationally infeasible, and (ii) associated methods to establish individual location-choice sets for this context have not yet been developed. We address this gap by framing location choice set formation as a graph-based node classification problem. Using Graph Neural Networks (GNNs), we predict, for each individual, the probability of visiting different locations across a transport network graph enriched with point-of-interest, land-use, and individual socio-demographic features. We then demonstrate two different sampling strategies to generate choice set from the predicted node probabilities. We validate our approach on a synthetic case-study as well as real-world historic travel data from the Geneva public transport network, showing that GNNs can recover behaviourally plausible location choice sets while remaining computationally tractable.

ABSTRACT. Agent-based transport simulations typically rely on a synthetic population and their travel demand over an average day. One important part of the travel demand generation is the assignment of secondary activity locations within trip chains. This work proposes an adapted multinomial logit framework for secondary location choice that jointly selects mode and location through importance sampling. The model incorporates location attractiveness from employment counts and distance decay parameters estimated from a national travel survey. Candidate locations are filtered by purpose and are sampled within spatially adaptive buffers, with a mixed local/global sampling scheme to preserve long-distance trip representation. The mode share is calibrated through an iterative ratio adjustment technique. Evaluated against the travel survey distance distributions, the proposed model generally reproduces similar distance distributions up to the 95th percentile of distances.

ABSTRACT. Traffic speed and ride-hailing demand are closely intertwined in dense urban areas, yet they are almost always predicted by separate, unconnected models. This paper presents STT-CTA (Spatial-Temporal Transformer with Cross-Task Attention), a single architecture that jointly predicts hourly traffic speed and ride-hailing demand on a regular urban grid using factorized spatial-temporal self-attention and a learnable-gated cross-task attention module that enables selective information exchange between tasks. We evaluate the model on one full year of data from Chicago's Loop district, combining traffic sensor readings, trip records, weather observations, and public holidays into a shared input representation. The key result is an \emph{asymmetric directional coupling}: letting the demand branch attend to speed signals reduces demand MAPE by 2.2 percentage points and MAE by up to 8.4\% at a 6-hour horizon, whereas feeding demand information back into the speed branch degrades speed accuracy. This asymmetry supports a hybrid deployment strategy and carries practical implications for the design of multi-task urban transportation forecasting systems, where different tasks may benefit uneq

ABSTRACT. Discrete choice model specification is a time-consuming task for modellers, who often specify and estimate multiple models while balancing fit, parsimony, and behavioural plausibility. We present Delphos, an Reinforcement learning agent that learns transferable specification strategies across transport choice datasets. Delphos frames specification as sequential decision-making problem in which it applies a sequence of modelling actions and receives rewards from an estimation environment based on goodness-of-fit and convergence. To transfer modelling decisions across datasets with different variable sets, we use a DeepSet-Q architecture that encodes specifications as sets of modelling terms and conditions decisions on a dataset context vector. This enables a shared policy to be trained in a multitask setting and to generalise its recommendations to unseen but related datasets. Delphos is trained on three mode choice datasets and increasingly proposes specifications that outperform the baselines over training. When then applied to an unseen dataset, the trained agent specifies competitive and behaviourally plausible models in less than 10 minutes on a CPU.

ABSTRACT. This paper proposes symbolic regression (SR) as an interpretable, data-driven approach to utility specification in discrete choice models. The aim is to address the limitations of traditional manual specification search and the lack of interpretability in common machine learning methods. The methodology formulates the systematic utility difference as an unknown function learned via SR, optimised using a logit negative log likelihood, ensuring consistency with random utility theory. SR generates a Pareto frontier of candidate specifications, balancing goodness-of-fit and complexity. Using synthetic binary choice data, we evaluate the ability of SR to recover known functional forms under varying search spaces, including missing transformations, irrelevant features, and noise variables. The results show that SR consistently recovers the true specification, reconstructs nonlinear effects endogenously, and remains robust to irrelevant inputs. These findings demonstrate the potential of SR to support systematic, interpretable specification discovery in choice modelling.

ABSTRACT. Traditional transit network design forces a choice between the structural rigidity of discrete graphs and the oversimplified approach of density-based continuous models. This paper bridges that divide by introducing a novel, corridor-based continuum framework that endogenously optimizes spatial coordinates and line-specific headways of transit networks. Driven by a stochastic routing model and a three-stage hybrid optimization strategy, the framework navigates a highly non convex design space while enforcing vehicle capacity at critical transfer nodes.

Our findings highlight a consistent design pattern: while homogeneous layouts provide good strategic guidance, introducing geometric and frequency heterogeneity balances trade-offs between capacity utilization and accessibility across the network under spatially heterogeneous demand. This fine-tuning improves the efficiency of service allocation, enabling operators to reduce costs by over 13% under centralized heterogeneous demand scenarios, resulting in substantial operational efficiency without degrading the passenger experience.

ABSTRACT. This paper proposes a mixed integer linear programming (MILP) framework for timetable optimization in mixed transit networks combining schedule-based and frequency-based services. The model incorporates user heterogeneity, distinguishing between regular and vulnerable users, and captures differences in passenger arrival behaviour at initial and subsequent stations. The approach is applied to the Copenhagen network across nine scenarios with varying demand composition and cost weights. Results show that timetable decisions are mainly driven by trade-offs between operational and passenger costs, while demand heterogeneity has limited impact on overall performance. However, vulnerable users consistently experience higher generalized costs, highlighting persistent equity gaps in transit systems.

ABSTRACT. Persistent road-network monitoring with multiple UAVs requires continuously refreshing time-sensitive traffic knowledge under battery constraints. Unlike one-shot coverage, monitoring quality depends on both where and when UAVs revisit road segments, making long-term coordination a central challenge. In this paper, we formulate persistent multi-UAV road-network monitoring as an asynchronous graph decision problem with temporally decaying road knowledge. We develop a decentralized monitoring framework in which UAVs repeatedly select road segments to maintain network-wide awareness while operating under charging constraints. To improve decision quality without sacrificing scalability, we combine learned decentralized control based on multi-agent proximal policy optimization (MAPPO) with short-horizon planning based on predictor + upper confidence bounds for trees (PUCT) search. The framework also operates in the presence of background delivery UAV traffic, whose motion affects coordination in the shared road network. Experiments in an urban road-network environment show that the framework provides an effective and practical solution for persistent monitoring and improves decentralized monitoring performance in dynamic road networks.

ABSTRACT. This study presents a robust two-echelon crowdsourced pickup and delivery problem with time windows (R2E-CPDPTW) to optimize routing and order allocation between first-echelon vehicles and second-echelon crowdsourced and contracted vehicles under travel time uncertainty. Formulated as a route-based set-partitioning model, it employs a cardinality-constrained uncertainty set to characterize travel time variations and preserve strict temporal synchronization between the two echelons. To solve this complex model to optimality, a customized branch-price-and-cut (BPC) exact algorithm is developed, where the pricing subproblem is solved as a robust shortest path problem with resource constraints. Computational results demonstrate the algorithm's efficiency in yielding exact solutions and highlight the critical trade-offs between system robustness and operational costs.

ABSTRACT. The rapid expansion of last-mile delivery has reshaped urban logistics, yet how delivery flows vary with distance remains poorly understood. Using approximately 48 million delivery records from multiple platforms across 12 cities on four continents, we provide the first large-scale, multi-city empirical estimation of flow-distance relationships in last-mile delivery. Semi-parametric spline and Gaussian process regression models reveal a universal exponential decay of express delivery flows with distance, while scheduled parcel deliveries exhibit near-uniform distance distributions. We further show that delivery fee discounts increase average ordering distances by 14\%, with lower-income customers responding most strongly. These findings demonstrate that the spatial and environmental externalities of urban delivery are behaviorally responsive to pricing strategies, with significant implications for sustainable urban logistics policy.

ABSTRACT. The online meal delivery services base speed and convenience as central service attributes, but this emphasis on fast delivery may compromise rider safety. Existing research has mainly focused on supply-side factors, such as rider behavior, to address the issue of safety, while little is known about whether consumers are willing to support safer delivery practices. This study addresses that gap through a stated-preference experiment with consumers of online food delivery services in Amsterdam and Copenhagen. Respondents chose between labelled cuisine alternatives that varied in delivery time, meal cost, delivery cost, and rider safety risk. A hybrid latent class choice model with random error terms was estimated to capture class-based heterogeneity and unobserved preferences. The results reveal two distinct consumer segments. One group is largely indifferent to rider safety, whereas the other is strongly safety-sensitive and willing to pay more or wait longer for safer delivery. These findings suggest that consumer-facing safety information in ordering apps could complement platform and regulatory interventions to improve the safety of delivery riders.

ABSTRACT. Promotion and assessment of sustainable and active mobility are increasingly widespread. Recent studies explore the role of electric-assisted bicycles (e-bikes) in travel behaviour, identifying motivation, substitution effects in other transport modes and changes at the individual level. However, evidence from longitudinal methods has been limited, and the use of either quantitative or qualitative approaches creates a gap between statistical generalisation and contextual understanding. This study employs a mixed-method approach to understanding the individual behaviour change process of e-bike adopters. Analyses are based on data from four waves of an online panel survey, semi-structured interviews conducted before and after with individuals who reported intention and then either adopted or did not adopt an e-bike, and approximately 9 months of mobility diaries from a mobile app. The study provides a comprehensive understanding of how, why, and under what conditions new technology adoption, such as e‑bikes, can support modal shifts.

ABSTRACT. Driven by advances in information technology, the sharing economy has rapidly expanded in the transportation sector. New demand has been identified and addressed by modes such as ride-sourcing and car-sharing, while travelers and the market remain in transition. This study explores travelers' mode choice preferences under the competition between emerging and conventional modes. A stated choice experiment was conducted in Northeast China, resulting in 337 valid responses, and a random parameter error component model was estimated. The results indicate that emerging and conventional modes form distinct groups with shared unobserved components. More importantly, their relationship is condition-dependent rather than uniform. Conventional modes dominate under structured conditions, while emerging modes become more competitive in flexible contexts. This suggests that emerging modes act as conditional competitors rather than direct substitutes. Competition across modes is asymmetrical, and socio-demographic factors such as income and familiarity with ride-sourcing further shape mode preferences.

ABSTRACT. Ride-hailing systems often offer both private hailing and pooling ride services, and platform performance depends on both how drivers are allocated across these services and how each service is priced. However, most existing studies treat these decisions separately or focus on a single service type, either private hailing or pooling. This paper proposes a multi-agent reinforcement learning framework to jointly optimize pricing and service type in ride-hailing systems under endogenous rider choice. Specifically, we formulate the problem as a multi-agent Markov Decision Process and train a pricing agent together with a service type agent using Proximal Policy Optimization in a realistic simulation environment. To ensure scalability across different fleet sizes, we employ a parameter-sharing architecture among service agents. Experiments on a realistic Delft road network demonstrate that the proposed framework learns stable and effective policies, consistently outperforming several rule-based baselines. The results highlight the substantial benefits of jointly optimizing pricing and service type in ride-hailing systems.

ABSTRACT. Commuters often persist with their current travel mode even when competing alternatives (e.g., robotaxis) become available. Whether this persistence reflects rational evaluation of service attributes or behavioral habit has been difficult to establish, because most studies do not vary the contextual conditions under which choices are made. This paper compares choices across habitual and new commuting contexts among 748 Shanghai commuters (410 car, 338 metro) using mixed logit and hybrid choice models. The results show that habits affect robotaxi adoption through two channels: they increase preference for the current mode in familiar settings and reduce responsiveness to competing service improvements, with both effects substantially stronger for car commuters. Car users required roughly three times the cost or time advantage needed to attract metro users. These findings suggest that robotaxi adoption forecasts based solely on service competitiveness will overestimate switching when established commuting routines remain in place.

ABSTRACT. Within this study, we estimate a Cross-Nested Logit (CNL) model incorporating ride-pooling and other shared modes using combined revealed and stated preference data. The CNL allows to captures complex substitution patterns that cannot be represented by simpler MNL or NL models. Results show that ride-pooling correlates with multiple nests and that behavior differs significantly between frequent and seldom users. Frequent users (using ride-pooling at least once a month) would rather replace public transport and cycling, while seldom users would shift to other shared modes. This could provide evidence that frequent users already incorpo-rated ride-pooling in daily travel routines while seldom users are still in the exploring phase. Overall, the study highlights the importance of accounting for behavioral heterogeneity and target-group specific analysis, when investigating new modes or setting up models to support decision-makers with scenario analyses.

ABSTRACT. Existing Mobility-as-a-Service (MaaS) simulation studies often lack full integration between dynamic demand generation and multimodal supply coordination, limiting their ability to capture realistic demand-supply interactions across transport modes. This study presents an integrated simulation framework that combines activity-based travel demand generation using SimMobility, mesoscopic traffic simulation with Aimsun, and a centralized MaaS controller to coordinate multimodal services. The framework is implemented on a virtual urban network modeled after Singapore. Experiments indicate that MaaS subscription offering discounted public transport fares shifts users toward public transport, reduces overall road usage, and improves Mobility-on-Demand service performance. At the same time, public transport systems accommodate the increased demand without loss of service quality. The proposed framework captures key cross-modal interactions and provides a practical tool for evaluating MaaS policies under realistic urban conditions.

ABSTRACT. This paper investigates how eye-tracking data can help explain why interventions work or don’t. Using an immersive virtual reality experiment in which participants repeatedly chose between taxi and bus, we compare a series of multinomial logit (MNL) and Decision Field Theory (DFT) models under static and dynamic specifications, with and without gaze-based information. The results show that eye-tracking improves behavioural interpretation in two ways. First, it helps distinguish which specific intervention elements entered the decision process, showing that not all parts of a nudge are equally effective. Second, when incorporated into DFT, gaze data provide a direct link between observed attention and preference formation. Across models, the dynamic structure better captures repeated decision making, and DFT offers a modest but consistent improvement in model fit. The findings provide useful guidance for policy makers by showing how intervention design can be improved through attention-based behavioural insights.

ABSTRACT. Pedestrian non-compliance at signalized intersections is a major safety concern in urban areas, particularly in locations with high pedestrian demand and complex traffic conditions. This study investigates the factors influencing pedestrian signal violations using video-based data collected at a signalized intersection in Athens, Greece. Pedestrian trajectories, vehicle trajec-tories, signal status, and interaction variables were extracted using a computer vision pipeline. Statistical analysis, including point-biserial correlation, binary logistic regression, and Random Forest classification, was applied to examine the relationship between traffic conditions and pedestrian compliance behaviour and to predict illegal crossings. The results indicate that pe-destrian violations are strongly associated with traffic conditions such as vehicle speed, Time-to-Collision (TTC), signal phase, and vehicle presence. The predictive models achieved high classification accuracy, indicating that pedestrian non-compliance is not random but influ-enced by observable traffic and interaction conditions. The findings support proactive road safety analysis and the development of predictive safety assessment tools.

ABSTRACT. Most methods for generating synthetic populations rely on cross-sectional snapshots or pseudo-panels, which do not track individuals consistently over time. This paper proposes a general framework for constructing synthetic populations whose panel structure is specified by design. Individuals are represented through life-based trajectories defined independently of calendar time, and a deterministic mapping recovers their state at any time t, allowing the reconstruction of panel data and population distributions at arbitrary points in time. The framework is model-agnostic and enforces internal consistency through structural constraints embedded in the life representation.

We further introduce a Bayesian updating mechanism that incorporates information from observed cross-sectional datasets. When data are available, the synthetic population is sampled from the posterior distribution, combining prior knowledge with the evidence contained in the observations. This allows cross-sectional information, such as census data, to inform the generation of coherent longitudinal populations.

ABSTRACT. Whereas synthetic populations constitute a fundamental component of agent-based simulation models, synthetic population forecasting seeks to generate such populations at future time horizons, enabling the simulation of hypothetical scenarios with representative agents. Existing approaches typically rely on static, single-year snapshots augmented with external data sources. In this work, we propose a novel framework for short-term forecasting that captures temporal trends directly from multi-year data, eliminating the need for external inputs. Our approach combines a projected Bayesian Network with a projective Iterative Proportional Fitting procedure. For each component, we perform attribute-level hyperparameter selection to differentiate stable relationships from those exhibiting meaningful temporal dynamics. Experimental results across diverse scenarios including varying training sample sizes, forecast horizons (one to five years), and attribute configurations, demonstrate that the proposed method yields improved distributional accuracy compared to static models, while maintaining comparable levels of realism. Additionally, we introduce a guardrail-based validation mechanism to detect and mitigate erroneous extrapolations.

ABSTRACT. Stress is a ubiquitous factor in modern life, yet its behavioural consequences in real-world mobility remain poorly understood. Understanding how affective states influence daily activity organisation is central to advancing behavioural modelling in transportation. While laboratory evidence suggests that stress promotes habitual decision-making and reduces behavioural flexibility, empirical validation in real-world mobility contexts remains limited. This paper summarises a first effort to develop an interpretable probabilistic modelling framework linking latent physiological stress to daily activity behaviour using multimodal wearable, GPS, mobility diary, and ecological momentary assessment data from 119 participants in a multi-day naturalistic experiment. Daily stress exposures are then tested, together with past behaviour, in explaining activity patterns the following day. Daily activity organisation is analysed along two complementary behavioural margins: (i) activity intensity, measured through the count of distinct activity types undertaken during a day, and (ii) participation in socially oriented activity types. Activity intensity is modelled using Poisson count regression with temporally ordered stress predictors, while participation behaviour is analysed using logistic regression. Results indicate systematic associations between prior stress exposure and subsequent behavioural organisation. Higher lagged stress is linked to measurable proportional changes in activity intensity and altered probabilities of social participation. Robustness analyses confirm that these findings are qualitatively stable across alternative stress constructions and heterogeneous across population subgroups. The study empirically demonstrates the importance of affective dynamics in daily-activity decision making from a naturalistic perspective.

ABSTRACT. Bus operators are increasingly seeking ways of electrifying their fleets, which is motivated by a combination of policy (decarbonization, emission standards) and economic (opera-tional efficiency) factors. For example, in London there is now an explicit objective set by the Mayor of London to transition to a fully electric bus fleet by 2034. At the same time, the dynamic, spatially disperse but predictable nature of bus services (routes, timetables) suggests their applicability as a potential source of flexible electricity storage and supply for the grid, via adjustable charging as well as discharging (vehicle-to-grid, V2G) capabilities. To-date, simulation-based analyses have demonstrated the potential for bus fleets to support electricity grid operations, for instance by enabling more volatile renewable energy genera-tion into the grid. The current paper advances the modelling of such benefits, by present-ing a modelling framework developed using Transport for London (TfL) working timeta-bles to map vehicle duty cycles at a 15-minute resolution. The approach defines bus archetypes using battery capacity and energy consumption, with load factors applied to account for passenger occupancy, auxiliary loads and time-of-day variations. State-of-charge dynamics were simulated for bus archetypes operating according to a real-world working timetable, with rules to safeguarded service reliability through departure floors, safety margins and export scheduling windows aligned with UK energy demand peaks. The empirical case study, based on Edgware Depot in London, indicated that fleet operating from that bus de-pot could shift up to 6,455 kWh in a single day, equivalent to the daily consumption of 650 UK households. Exports were concentrated in the evening super-peak, though some capaci-ty was available during the day due to school services and certain buses dwelling in the de-pot during the interpeak period. Route and vehicle level analysis revealed substantial heter-ogeneity, with some buses unable to participate due to continuous duties, while others of-fered disproportionate contributions.

ABSTRACT. This paper proposes the Markov perturbed utility route choice (M-PURC) model. The model is based on a generalized entropy over link flows, constructed at each node as the perspective of a convex node perturbation with respect to the node outflow. The convex conjugate of the generalized entropy defines a surplus function whose gradient yields the optimal link flows. Lagrangian duality decomposes the surplus maximization into node-level perturbed utility problems coupled through a Bellman equation, yielding a Markovian representation of optimal behavior. The framework accommodates the recursive logit and the NGEV model as special cases. A separable specification, in which the node perturbation is the sum of link perturbations, yields a link-by-link probability-recovery formula and endogenous sparsity: dominated links receive exactly zero probability without choice set specification. We estimate the model via a Fenchel--Young loss that is convex in parameters, remains finite at corner solutions, and reduces to the recursive-logit likelihood as a special case.

ABSTRACT. This paper adapts the Free Energy Principle (FEP) to route choice under uncertainty and examines whether an active inference account can explain travel learning better than conventional reinforcement learning (RL). We develop a route choice model in which travellers update beliefs about hidden traffic conditions and select routes by minimising expected free energy, thereby jointly capturing preferences, uncertainty, and exploration. The model is estimated in a hierarchical Bayesian framework and applied to two complementary experimental datasets: an incentive-compatible driving simulator study and a laboratory route choice study with information manipulations. Across both datasets, the FEP model provides a better account of observed behaviour than RL, especially for participants showing exploratory or belief-adaptive responses. The results suggest that active inference offers a promising neurocomputational foundation for richer models of adaptive travel behaviour.

ABSTRACT. Docked bike-sharing systems (DBS) play a pivotal role in promoting sustainable urban mobility by facilitating active travel and reducing motorized congestion. However, their long-term efficiency relies on each station’s ability to self-balance under temporally and spatially uneven demand. This study investigates the intrinsic self-balancing behavior of Barcelona’s Bicing network throughout 2024, encompassing 412 stations at a daily resolution. We introduce a station self-balancing index ($SSI$)—a normalized indicator capturing day-to-day dock-availability variability—and employ a LightGBM–SHAP framework to quantify nonlinear associations among spatial, climatic, and temporal factors. Results show that spatial attributes, particularly residential and recreational points of interest (POIs) and building density, account for roughly two-thirds of the model-explained variance in $SSI$. SHAP interaction analyses further reveal that weekday work-oriented flows are associated with higher imbalance levels, whereas leisure-oriented urban morphologies help stabilize weekend operations. Overall, the findings provide interpretable, data-driven insights for enhancing DBS planning and operations, while offering a generalizable framework for assessing the self-regulatory capacity of urban micromobility systems.

ABSTRACT. Dense suburban railway systems are subject to various types of hazards whose effects on supply and demand interact at short time scales: dynamic interactions. The combination of hazards and dynamic interactions may lead delays to propagate and worsen, which significantly deteriorates the system's performance. As such, it is crucial to simulate these effects to design robust schedules. However, existing simulators for dense suburban railway systems do not fully include demand-to-supply and supply-to-demand dynamic interactions. In this paper, a proof of concept simulator is presented to show the importance of considering demand-to-supply and supply-to-demand dynamic interactions. Compared to a benchmark simulator that neglects these dynamic interactions, on the presented case study, our simulator is able to highlight additional phenomena resulting from demand-to-supply and supply-to-demand dynamic interactions. In practice, the implementation of all dynamic interactions in a ready-to-use simulator will significantly improve future scheduling. This work is a first step toward this goal in which a proof of concept is provided.

ABSTRACT. Flexible transit services are widely used as many-to-one feeders in rural and suburban areas with intermediate demand, yet their flexibility often induces substantial in-vehicle travel-time penalties due to cumulative deviations. This study investigates whether modular vehicles enabling in-vehicle transfers can support partial express operations that mitigate such penalties without external transfers. We develop an analytical model for a zone-based feeder system with checkpoints, capturing re-partitioning of service zones as express units depart. Performance is evaluated using an integrated generalized-cost framework capturing operator and user costs, including walking and riding. Results across a structured design space of demand densities and service-area geometries indicate that modular express operations can reduce total generalized cost in various configurations, primarily driven by peak riding-cost reductions of approximately 19–22%. The findings highlight the potential of modular coupling and in-vehicle transfers as an operational enhancement for flexible feeder systems under intermediate demand conditions.

ABSTRACT. The cycling gender gap and bicycle ownership-usage mismatch represent critical bottlenecks, particularly in France’s sustainable mobility transition, as nearly a third of privately-owned bicycles are unused. This study utilizes a nationally representative sample of the 2019 French National Travel Survey (4817 males; 3494 female bike owners) to investigate behavioral heterogeneities. Partial Proportional Odds models estimate cycling frequency, controlling for age, commuting, residential density, and car ownership. Males exhibit baseline odds of cycling frequency 3 times higher than females across all thresholds. However, distinct gendered drivers emerge: while complex household structures correlate with increased frequency for men, the presence of children induces cycling frequency for females. Higher income decreases the odds of frequent cycling by 25% for males and 35% for females. Observably, e-bike adoption strongly increases cycling frequency. These suggest that the e-bike is a vital bridge to overcome car usage inelasticity and mitigate household structural barriers, toward gender-equitable cycling.

ABSTRACT. This paper proposes a city-scale operational framework for station-based micromobility rebalancing that combines censored-demand reconstruction with hybrid truck coordination. The method introduces Shortage-Aware Tobit Attention (SATA) to recover latent demand when low bike inventory suppresses observed departures, using inventory states, shortage-aware attention penalties, and external taxi signals. The reconstructed demand is converted into shortage and surplus candidates and optimized through a hybrid service-MPC that coordinates deterministic operator trucks with stochastic crowdsourced delivery vehicles under timing, detour, capacity, and compatibility constraints. An incentivization model estimates crowd acceptance and feeds expected participation into the control layer. Compared with the operator-only baseline, the full hybrid stack improves service rate by up to 31.5% in Chicago stress scenarios and 13.2% in Lyon crowd-advantage settings. Operator profit increases by 36.3%.

ABSTRACT. Rather than assuming fully compensatory choice among many physical cycling routes based on cumulative link utilities, this study considers an alternative formulation in which cyclists make decisions at an abstract level between "labeled" route types given bounded rationality. A route type can be understood as a representation of routes with certain favorable characteristics. Although realizations of route types can be generated using shortest path search for each trip, they may not always be sufficiently distinct from the shortest route to be perceived as viable alternatives. To account for implicit perceived availability, we formulate the route type choice problem via a constrained multinomial logit model (CMNL). Using GPS traces from Zurich, we demonstrate the approach and identify attribute thresholds that influence the perceived availability of different route types. The proposed framework provides a more interpretable representation of cycling route choice and improved behavioral consistency with choice set generation.

ABSTRACT. Providing public transport in demand-sparse rural areas means balancing operational efficiency with equitable service for mobility-deprived users. This paper introduces a Needs-Based Dial-a-Ride Problem (NB-DARP) that integrates behavior models with vehicle routing optimization. Using panel data from rural Japan, we adapt multi-day activity models into a scheduling constraint generation framework to assess user flexibility, and predict availability. Solving the NB-DARP with an Adaptive Large Neighborhood Search (ALNS) metaheuristic, we demonstrate that it outperforms traditional "first come, first served" and ridership-maximizing policies. The proposed framework increases overall service rates and ensures equitable transit access without requiring substantial temporal shifts from users.

ABSTRACT. The development of Level 2 driving systems complicates the interpretation of Naturalistic Driving Study (NDS) data, as behavioral modeling requires accurate identification of the operational mode. While longitudinal control has already been studied, lateral behavior remains underexplored, particularly in the frequency domain. This study proposes a classification framework to distinguish between SAE Level 0 and Level 2 driving using the Power Spectral Density (PSD) of the vehicle's lateral position. High-precision trajectory data (10 Hz) from six instrumented vehicles were collected on the I-24 MOTION testbed, and the signals were split into short driving episodes. We extract five spectral metrics, band power, slope, centroid, band power ratio, and entropy, to capture distinct control signatures. Statistical analysis confirms significant differences between driving modes across all examined time horizons (15-90 s). A logistic regression model achieves a peak accuracy of 92% at a 60-second observation window, identifying this as the optimal duration. These results demonstrate that frequency-domain methods effectively separate machine control from human steering, providing a foundational tool for large-scale safety analysis and automated dataset labeling.

ABSTRACT. An essential component of any accessibility measure is an impedance function capturing the frictional impact of travel on activity participation. Such functions are commonly calibrated using the marginal distribution of the impedance variable, usually distance or time. However, this approach lacks a clear theoretical basis, and is affected by bias related to the fact that shorter trips have fewer available destinations than longer trips, meaning that trips with low impedance occur less frequently. This bias may explain the observed bell-shaped survival functions for impedance, which are often attributed to frictionless responses at lower impedance values. This paper develops a theoretical foundation for calibration to marginal impedance distributions, proposing a power function correction to isolate the bias from the impedance function during estimation. Empirical results show that applying near‑linear correction terms leads to better-fitting distributions. The resultant impedance functions exhibit steeper decay, carrying important implications for accessibility measurement and policy interpretation.

ABSTRACT. This study develops an integrated optimization framework for mitigating crowding in urban rail transit systems through the joint design of train scheduling, dynamic pricing, passenger arrival time shifting, and passenger flow control under demand uncertainty. Unlike conventional studies that treat time-varying demand as exogenous, we explicitly model fare-dependent passenger arrival time choice. The problem is formulated as a stochastic program that minimizes total passengers' travel time subject to train capacity limits, service priority for reserved passengers, and a government subsidy budget. To solve the model, we develop an improved Benders decomposition method further enhanced by machine learning. Numerical results show that the proposed approach effectively reduces passengers' waiting time and improves computational efficiency.

ABSTRACT. This paper presents a decision-support framework for designing integrated fare systems across fixed-route and on-demand public transport. Using one year of operational and ridership data from Ingolstadt (Germany), we generate a set of ring- and cell-based zoning schemes and calibrate fares to preserve current single-ticket revenue. For the evaluation, we focus on the trade-off between an easy-to-understand tariff system, captured by structural complexity, and price-distance fairness (Gini index). Fare-system performance is first assessed separately for fixed-route and on-demand services. In a second step, Pareto-efficient joint fare systems are identified and ranked by their weighted Euclidean distance to an ideal point representing minimum complexity and maximum fairness. Findings show that moderate zoning granularity can improve both simplicity and price-distance fairness. Overall, the framework provides transport authorities a decision-support tool for selecting user-oriented integrated fare systems.

ABSTRACT. European long-distance rail is fragmented into national-oriented networks, creating impedance for international travellers. This study analyses cross-border impedance across Europe via transfer counts, using open General Transport Feed Specification (GTFS) data to construct a P-space graph of the European network. We introduce the maximal reach and country adjacency degree concepts to present the rail network structure in terms of concentric neighbouring relations between countries. We then characterised the countries in terms of connectivity to urban population, and number of transfers between country pairs. Our results reveal a clear core-periphery structure, where Western and Central European countries achieve high connectivity within 1–2 transfers, while peripheral countries require additional transfers. Directional heterogeneity is evident at borders, characterised by clusters of well-connected countries and poor core-periphery links. On average, transfer counts have a 1:1 relation with number of borders traversed, and high impedance locations generally agree with previous studies on cross-border travel speeds.

ABSTRACT. Metro services shape urban built environments, but comparative macro‑scale studies across de-veloping regions remain scarce. This research evaluates 64 cities (42 Chinese, 22 Latin Ameri-can) using hierarchical and k‑means clustering across three scenarios: density of metro services, metro area coverage (400 m buffer), and network extensiveness per population. Four distinct city‑metro interaction typologies emerge. Mega Chinese cities show high node‑high place pat-terns (Cluster C3), while medium‑sized Latin American cities exhibit intermediate node‑place balance (Cluster C4) – a previously undocumented potential for pedestrian‑oriented TOD. The study demonstrates that walkability proxies (400 m buffer) can be operationalized at macro scale, and that compact urban form is a necessary condition for effective TOD, even with limited metro coverage. Findings provide actionable typologies for context‑specific infrastructure and pedestri-an policies.

ABSTRACT. This paper applies a quantitative spatial model (QSM) to the Ceneri Base Tunnel in Ticino, Switzerland, producing general equilibrium predictions of how a major rail investment reshapes the spatial economy. The model adapts the Hörcher and Graham (2025) framework, which extends the standard QSM with a dual budget constraint that separates monetary and time costs of commuting, yielding an endogenous, pair-specific marginal value of time. Because spatially disaggregated floorspace price data is unavailable for the study area, we reverse the standard QSM inversion procedure: prices are recovered from observed floorspace stock using endogenous aggregates from the commuting block — an approach that, to our knowledge, has not been implemented in the QSM literature. Applied to 195 zones in the Bellinzona–Lugano corridor, the model predicts employment concentration in Lugano (+5%), residential decentralisation towards Bellinzona (+2%), and a productivity gain in Lugano (+0.3%) driven by agglomeration externalities. Wages decline across both agglomerations as effective labour supply rises. The average value of time experienced by commuters falls from 17.7 to 17.4 CHF/hr, yet the structural welfare measure registers a positive gain (+0.08%) — a divergence that connects to the theoretical literature on the value of travel time savings and has implications for how transport investments are evaluated.

ABSTRACT. Autonomous vehicles are widely expected to generate substantial social, environmental, and economic benefits. However, these anticipated effects are still largely derived from simulations, models, and forecasts, which have so far focused predominantly on dense urban settings. This emphasis leaves an important gap, as autonomous vehicles deployed as a shared and pooled mobility service may be particularly valuable for improving accessibility and mobility in suburban and rural areas, where such benefits have rarely been assessed. Building on empirical mobile-phone origin–destination flows, this study presents the first agent-based simulation of an autonomous ridepooling system at a nationwide scale. The model represents a service operating across all of Germany, serving nearly 12 million daily rides with a fleet of 300,000 vehicles, and evaluates its implications for users, operators, and society. The results indicate a pooling rate of 77\%, a mean vehicle occupancy of 1.67, and an average waiting time of 10 minutes. In addition, the system is estimated to reduce external transport costs by 3.6 billion euros annually. These findings suggest that, if effectively implemented, autonomous ridepooling could deliver substantial individual, societal, and environmental benefits not only in cities but across an entire country.

ABSTRACT. We examine whether route-choice equilibria remain stable once connected autonomous vehicles (CAVs) can coordinate their decisions. In classical traffic assignment, a user equilibrium is interpreted as a Wardrop or Nash-type equilibrium where no traveler can improve by changing route alone. We show that this will be no longer sufficient when autonomous vehicles can communicate and deviate jointly. We formulate route-switching coalitions (clubs), define their internal and external stability, and demonstrate the formation mechanism in a microscopic mixed-traffic network with adaptive traffic signals. Starting from a user equilibrium, a three-vehicle CAV club can profitably deviate to another route, reduce each member's travel time, worsen conditions for excluded users, and increase the total travel time in the system. The results imply that CAV coordination may generate inequitable and strategically unstable traffic states on public road networks.

ABSTRACT. Over the past fifteen years, collaborative practices have expanded rapidly, driven by digital technologies and changes in urban lifestyles, they have contributed to the rise of the sharing economy, of which home-sharing and carpooling are two examples that are particularly widespread among students. This study analyzes the influence of student shared housing on the use of carpooling for home–campus commuting. While these two practices may be complementary, their relationship remains theoretically ambiguous due to coordination constraints, rebound effects, or moral licensing mechanisms. The analysis relies on original data from the MobiCampus-UdL survey conducted among around 9,000 students in the Lyon metropolitan area, using a recursive bivariate discrete choice model based on a copula approach to address the endogeneity. Results show a positive causal effect of shared housing on carpooling, but also reveal strong self-selection, suggesting shared housing alone is insufficient to generalize carpooling.

ABSTRACT. Peer-to-peer (P2P) car-sharing enables privately owned vehicles to be temporarily shared through digital platforms, yet its realistic urban potential remains insufficiently understood. This study applies synthetic populations to assess P2P car-sharing potential. Daily activity–travel patterns are analyzed to extract demand and supply. A two-step assignment process identifies spatially and temporally feasible matches and determines compatible assignments under operational constraints. The framework is applied to French cities, Dunkirk, Nantes, and Paris, representing diverse urban contexts. Results show that walking distance strongly influences matching levels, while temporal patterns are consistent across cities. A significant share of individuals have no feasible match, revealing spatial and temporal imbalances, while a small subset accounts for most interactions. Findings indicate that P2P car-sharing shows the potential to reduce car ownership and parking demand, and support public transport in low-access areas. The framework provides a scalable tool for assessing operational feasibility at the urban scale.

ABSTRACT. Connected and autonomous vehicles depend on inter-vehicular communications to perform cooperative driving tasks. Hardware failures and malicious attacks can cause communication disruptions, potentially affecting traffic efficiency and safety. In this paper, we proposed a multi-anticipative IDM with a range-based communication topology under delay. Information from multiple predecessor vehicles weighted through an exponential distribution incorporating cumulative distance and signal decay parameter. We derived stability conditions under communication delay to evaluate model stability under different parameter variations. The critical communication delay $\tau_c(v_e)$ is derived, beyond which the platoon becomes unstable. Also, the model is evaluated under clustered communication failure, where consecutive vehicles lost communication ability due to a malicious attack. Simulations show improved oscillation dampening and reduced velocity error than baseline IDM. Safety impact evaluated over a range of failure durations through multiple runs of simulations.

ABSTRACT. Connected and Automated Vehicles (CAVs) operating on lane-free highways offer substantial gains in traffic efficiency. However, their inherent nonlinear dynamics and the presence of coupled, non-convex safety constraints present critical challenges to control design. Centralized Model Predictive Control (MPC) ensures safety, but suffers from scalability and communication limitations. To address these challenges, this paper investigates decentralized MPC (DMPC) for CAV coordination, focusing on iterative, non-cooperative algorithms, including Jacobi-type and Gauss-Seidel-type. A novel decoupling method is developed to transform nonconvex safety constraints into convex, locally enforceable constraints, inspired by buffered Voronoi cells. The simulation results show that the proposed DMPC algorithms achieve safe and efficient vehicle trajectories while substantially improving scalability, highlighting their potential for future lane-free CAV traffic systems. Ultimately, the results indicate that the most suitable decentralized control strategy depends on the desired trade-off between safety, performance, and computational efficiency.

ABSTRACT. In urban periodic deliveries, business customers exhibit heterogeneous preferences over price, visit frequency, timing, and time window length. Ignoring such heterogeneity leads to service designs that misalign with customer needs. We propose the choice-based periodic vehicle routing and pricing problem (CB-PVRP), which incorporates stochastic customer choices into joint optimization of service offering, pricing, and vehicle routing. We formulate CB-PVRP as a two-stage stochastic mixed-integer program. The first stage determines offerings and pricing to maximize expected profit. The second stage, which embeds a bilevel structure, decides daily routing subject to revealed choices. We capture preference heterogeneity using a multinomial logit model with segmentation. We apply sample average approximation to address stochasticity and propose linear best-response sets for the bilevel reformulation. Our tailored logic-based Benders decomposition algorithm outperforms a commercial solver on solution quality and speed. By incorporating customer choice behaviors, the provider achieves higher expected profitability while customers receive improved services.

ABSTRACT. The development of on-demand transportation and Modular Autonomous Vehicles (MAVs) presents opportunities for dynamic capacity adjustment. However, existing MAV routing models are primarily restricted to fixed transit corridors, limiting their full potential. To bridge this gap, this paper introduces a novel MAV routing problem with pickup and delivery in a fully flexible environment. We propose a model aimed at minimizing total travel costs and penalties for unserved requests, while explicitly incorporating the cost-saving platooning effects of docked MAVs. The problem is formulated as an arc-flow model and reformulated using Dantzig-Wolfe decomposition. To achieve optimal solutions, we develop a branch-and-price algorithm incorporating a labeling algorithm to solve the pricing subproblem. Numerical results show that our exact algorithm outperforms solver in terms of solution speed.

ABSTRACT. This study aims to provide a review of factors influencing long-distance travel by rail and aims to identify key research gaps, challenges, and needs. We specifically focus on long-distance trips between 100 and 1500 km made by both conventional and high-speed rail. For this, we first propose a conceptual model that allows for categorizing the wide range of factors discussed. Next, by reviewing 67 studies, we identify determinants in four directions: journey, spatial, socio-demographic, and socio-psychological characteristics. To support the review, a meta-analysis is conducted on travel fares and travel duration. Own-price elasticities range between -0.32 and -0.50 whereas own-time elasticities range between -0.34 and -1.53. Finally, we offer a research agenda that highlights current gaps and discusses potential future directions that allow for advancing our understanding of how and which determinants influence long-distance travel choices so that effective policies can be designed.

ABSTRACT. We consider pedestrians who move along a corridor and cross the road at one of several signalised crossings. Such movement is often hindered by waiting at crossings, reducing walkability. Unlike vehicular green waves, which require coordinated progression through sequential intersections, pedestrians need to cross only once and may do so at any encountered crossing along their path. We study signal coordination that maximises vehicular throughput while guaranteeing that pedestrians can cross within a bounded waiting time. Signal timings are optimised using a genetic algorithm, and vehicular performance is evaluated using a Link Transmission Model. Results for two- and three-crossing corridors show that small waiting allowances substantially increase throughput, with the largest marginal gains at low waiting values. Increasing waiting also changes how pedestrian service is coordinated, shifting from distributed coverage across crossings towards concentrated service at fewer locations and shorter cycle times, as temporal waiting increasingly substitutes for spatial coverage.

ABSTRACT. The widespread adoption of geolocation technologies has transformed human mobility analysis, offering an alternative to traditional Household Travel Surveys (HTS). However, these raw mobility datasets typically lack essential socio-demographic labels, making it difficult to assess their statistical representativeness. This study investigates the potential of transferring models trained to infer individual socio-demographic attributes from spatio-temporal trajectories to unlabeled target datasets, in order to reconstruct global population distributions. Using French and Swiss mobility datasets, we evaluated the robustness of this transfer approach. Our results demonstrate that model portability is challenged by dataset shifts. Cross-domain transfer severely degrades individual-level predictive performance. We study quantification as a post hoc correction for prior shift and show that Probabilistic Classify and Count (PCC) remains effective in recovering class prevalence under distributional shift. We further outline directions for learning domain-invariant representations and developing new alignment strategies to address covariate and concept shift, improving generalization across domains.

ABSTRACT. The influence of commuting on work–family conflict (WFC) and life satisfaction is rarely analysed within a unified framework. This study investigates the relationships between commuting, WFC, and life satisfaction using German data from the German Time Use Sur-vey (GTUS 2012/13) and the German Family Panel (pairfam 2019). Two complementary structural equation modelling (SEM) mediation analyses are conducted, each addressing a different outcome due to data constraints. Results from the GTUS analysis show that com-muting increases WFC indirectly through time-use mechanisms, particularly reduced leisure time, changes in personal care, and unpaid work, with clear gender differences in these pathways. The pairfam analysis further shows that commuting does not directly reduce life satisfaction but operates indirectly through WFC, while job and family satisfaction strongly enhance life satisfaction. Overall, commuting affects well-being primarily through indirect pathways involving time allocation and WFC rather than strong direct effects. The findings highlight the importance of reducing commuting distances, supporting flexible work arrangements, and addressing gendered divisions of care work, contributing to debates on the commuting paradox and transport–wellbeing linkages.

ABSTRACT. Pedestrian road-crossing is a safety-critical, time-pressured decision that produces both choices and response times, and recent work has analysed EEG correlates such as the centro-parietal positivity (CPP) in a separate, post-hoc step. We present a single-trial joint neurocognitive model that treats crossing choice, response time, and trial-wise CPP features as co-generated by shared latent evidence-accumulation dynamics. The model is fitted hierarchically to an EEG+behaviour pedestrian crossing dataset with repeated time-to-arrival manipulations, using simulation-based Bayesian inference. The joint formulation enables cross-modal prediction (early EEG windows sharpening predictions of imminent choice/RT; observed behaviour predicting CPP) and provides an explicit trial-alignment diagnostic via EEG--behaviour shuffling. Results show that one parameterisation can reproduce behavioural and neural patterns simultaneously and yields improved mechanistic interpretability compared to two-stage pipelines, supporting more robust prediction and model checking in road-user behaviour research.

ABSTRACT. This study proposes a novel interpretable clustering framework to identify traveler profiles that can support the design of sustainable transport policies. The approach introduces an unsupervised tree-structured clustering method that explicitly incorporates contextual feature subsets into the clustering process. By organizing features into predefined subsets and assigning them to successive layers of a tree, the method enables domain knowledge to guide how traveler profiles are constructed. At each node, the next clusters are generated using agglomerative hierarchical clustering with distance measures suited to numerical, categorical, or mixed data. A genetic algorithm is then used to explore and optimize the space of possible tree structures by maximizing a weighted average silhouette score. Applied to data from the Netherlands Mobility Panel, the framework produces traveler profiles defined by travel behavior and sociodemographic characteristics while maintaining interpretability in the clustering process.

ABSTRACT. Cyclist delays at intersections strongly affect perceived trip quality and network performance through interruptions, i.e. slowing, stopping, negotiating conflicts, and re‑accelerating. We aim to measure these delays consistently through GNSS trajectories by taking into account the cruising conditions of each situation and analysing factors affecting them. Steady-state cruising segments are identified, and a mixed‑effects model predicts cyclist‑specific cruising speeds based on infrastructure, slope, environment, and rider heterogeneity. Intersection delays are then defined as the ratio of observed crossing time to this personalised cruising-time baseline, with crossing distance extracted through a consistent geometric method. Results of a mixed-effects delay ratio model show that traffic signals, left turns, and peak periods are the dominant sources of delay, whereas right turns and off‑peak times reduce them. These findings point to clear intervention priorities: reducing signal‑related waiting through improved phase coordination, simplifying left‑turn movements, and minimising conflict points through improved intersection layouts.

ABSTRACT. This paper studies how congestion and spatial scope affect the evaluation of large-scale public transport investments in a quantitative spatial model. Using the planned cross-city railway tunnel in Budapest as a case study, we extend an urban general equilibrium model in two ways. First, we incorporate endogenous car travel times by estimating a route-based congestion function using origin--destination commuting data, free-flow travel times, employment distribution, and road capacity. Second, we expand the model’s geographical scope from the administrative area of the city to its entire conurbation, with hundreds of suburban settlements. The results show that accounting for congestion matters for policy evaluation: ignoring endogenous car travel times leads to an overestimation of aggregate welfare gains by 3.75%. We also find that restricting the model to Budapest’s administrative boundaries distorts counterfactual predictions, including floorspace prices, wages, and welfare. The findings highlight the importance of both congestion and functional urban boundaries in transport appraisal.

ABSTRACT. The integration of Vehicle-to-Grid (V2G) services into electric autonomous ride-sourcing fleets creates a fundamental trade-off between spatial displacement for mobility and temporal retention for energy arbitrage. This study investigates the stationary equilibrium of such large-scale fleets to analytically characterize the structural properties of the optimal fleet management strategy. We formulate a mean field game (MFG) framework under continuous energy constraints. We prove that the optimal policy exhibits a unique, spatially dependent threshold structure characterized by forced charging and V2G participation boundaries. To compute the equilibrium, we develop a Structure-Preserving Threshold Fictitious Play (SP-TFP) algorithm that ensures robust convergence by operating directly on the threshold manifold. Numerical results reveal that the equilibrium spontaneously organizes into Voronoi-like service cells. We further demonstrate that V2G infrastructure can either smooth spatial value disparities or exacerbate market polarization, depending on its deployment relative to demand hotspots. Ultimately, this study offers platform operators strategic insights into energy-spatial sorting and provides policymakers with actionable levers to enhance service equity and network robustness through targeted infrastructure planning.

ABSTRACT. This study examined the wider barrier effects generated by ground-level urban expressways and railways. To address spatial confounding, spatial dependence, and spatial spillover in a cross-sectional causal-inference design, this study developed a spatial causal-inference framework that combines a multi-treatment design with Double Machine Learning and Gaussian-process-based nuisance learning. Using Seoul’s ground-level transport networks and officially assessed land prices, the analysis found that the direct effects of transport infrastructure were mostly insignificant, whereas spillover effects were significant in most areas and reduced officially assessed land prices by an average of 11.12%. The results also showed that land-value declines were particularly pronounced in peripheral and relatively marginalized areas of Seoul, suggesting that transport infrastructure accelerates decline in already disadvantaged neighborhoods.

ABSTRACT. Influenced by sociocultural norms and ongoing safety shortcomings in urban transporta-tion systems, women frequently perceive a higher personal risk during routine travel. Building on the concept of the “pink tax” in consumer markets, this study introduces the notion of a transportation pink tax as a gendered safety-compensation burden, the addi-tional expected monetary cost that women incur when safety concerns prompt compensa-tory behaviors, such as choosing more expensive travel modes. Using a stated preference (SP) experiment conducted in Shanghai (N = 785; 391 females and 394 males), we exam-ine mode choice among metro, ride-hailing, and autonomous robotaxi services. Both mul-tinomial and mixed logit models are employed to estimate gender-differentiated prefer-ences and derive willingness-to-pay (WTP) for safety-related attributes. The results indi-cate that women place significantly higher marginal utility on safety-enhancing features, including female ride-hailing drivers, advanced in-vehicle security systems, and improved lighting, reflecting a greater safety-compensation premium in their utility functions. For example, women’s WTP for advanced security measures exceeds that of men by 3.90, 26.05, and 27.86 CNY for short-, medium-, and long-distance trips, respectively. Im-portantly, we show that this marginal safety premium translates into a tangible cost burden when women shift toward higher-cost modes under unsafe conditions. Policy scenario simulations indicate that enhancing metro safety, through measures such as shorter walk-ing distances, CCTV installation, increased security patrols, and improved lighting, could raise women’s metro usage and reduce their monthly travel expenditures by approximately 60–115 CNY, thereby alleviating the gendered safety-compensation burden. These results suggest that transportation gender inequities arise not from inherent preference differences, but from structural safety externalities that disproportionately impact women. Implement-ing safety-by-design principles in public transit and emerging autonomous mobility ser-vices can therefore lower gendered cost burdens and foster more equitable urban mobility.

ABSTRACT. Collecting behavioral data involves a fundamental tradeoff: richer data improves model estimation, but increases respondent burden and degrades response quality. Despite its importance, this tradeoff is rarely addressed in a unified framework linking survey design and behavioral inference. This study proposes a “missing-by-design” framework that treats partial observation not as a data limitation, but as a controllable design variable optimized prior to data collection. Rather than designing surveys for statistical representation, we design them for model estimation, explicitly aligning data collection with inferential objectives.

The framework formulates survey design as a bilevel optimization problem. At the upper level, a collection policy determines which optional questions to ask each respondent, subject to a constraint on respondent burden. At the lower level, a Bayesian inference procedure jointly estimates model parameters and imputes missing covariates via data augmentation, ensuring that design and inference are governed by a single probabilistic structure.

We demonstrate the framework using an MDCEV model of daily time allocation estimated on 42,199 respondents from the American Time Use Survey. Benchmark experiments under missing completely at random (MCAR) show that posterior standard deviations increase modestly with missingness—by 3.7% at 20% missingness and 10% at 40%—while preserving most behavioral conclusions (96.4% sign agreement at 20%). These results validate the inferential engine and establish a practical operating region.

Importantly, MCAR represents a lower bound on performance. By allocating survey effort based on inferential value, the proposed framework enables more efficient data collection, achieving lower respondent burden without compromising estimation quality.

ABSTRACT. Understanding factors influencing cycling safety is critical for promoting sustainable mo-bility. Cyclists require both objective safety (protection from crashes) and subjective safe-ty (feeling safe). While previous research has examined these separately, studies typically analyze only limited elements and focus on a single safety type at a time. This study inte-grates street-level imagery and spatial data to characterize how urban cycling contexts in-fluence both safety types. Using machine learning (XGBoost + SHAP), we analyze the im-pacts of 120 urban elements on 3,850 cycling crash locations in Berlin, Germany. Results show some elements have similar effects on both safety types, others affect only one type, and others have opposing effects. These findings highlight the importance of aligning ur-ban planning with both actual and perceived safety needs, as interventions targeting one type may inadvertently harm the other.

ABSTRACT. Growing concerns over climate change have accelerated transport electrification. Transport for London (TfL) commits to fully electrifying its bus fleet by 2034. While this transition brings challenges to grid infrastructures, simultaneously, it provides opportunities to integrate Vehicle-to-Grid (V2G) to reduce operational costs and enhance grid stability. This study develops a modelling framework using London bus open data to assess the potential of V2G under different charging and discharging cycles, state-of-charge (SoC) range, battery degradation, passenger load, and tariffs. Results show that maintaining SoC between 40–88% provides a balance between cost savings and V2G capacity. Wider SoC ranges reduce costs but accelerate battery ageing, limiting long-term benefits. Passenger load had little effect on V2G capacity (<5%) but significantly influenced operating costs. Fluctuating tariffs predictably affected costs but not V2G potential. Overall, transitioning from petrol to electric buses with V2G integration could deliver up to 98% cost savings over 14 years.

ABSTRACT. This study quantifies the economic potential for individual electric vehicle (EV) owners to optimize their charging and participate in Vehicle-to-Grid (V2G) under dynamic electricity pricing and data-driven mobility patterns. Using empirically derived vehicle usage profiles from the mobiTopp model and high-resolution day-ahead price data, charging decisions are modeled from a user-optimal perspective. Three strategies are compared: uncoordinated charging, price-optimized charging, and V2G-optimized charging. Results show that price-based optimization reduces total charging costs by up to 23.5%, while V2G further decreases average costs by over 90%, enabling negative net costs for a significant share of users. However, benefits vary considerably across vehicles. Additionally, price-driven charging leads to temporal concentration of demand and pronounced load peaks. These findings highlight both the economic potential as well as the emerging system-level challenges.

ABSTRACT. Most existing studies on bus transport systems treat service reliability and operational sustain- ability as separate objectives. This paper presents an integrated simulation and optimal control framework that jointly addresses both. We develop a real-time, event-based simulation environ- mentthatcaptureshuman-drivenbusbehaviorunderrealistictrafficconditions, passengerdemand, and operational disturbances, calibrated using open-source data. Building on this platform, we design and implement two complementary control strategies within a Model Predictive Control (MPC) framework: holding control at stops and speed limit regulation. The controllers are jointly optimized to regulate time headways while minimizing energy consumption. The framework is evaluated on five high-frequency bus lines in the Tel Aviv metropolitan area. Simulation results show that the combined strategy reduces headway deviations by over 60% while achieving ap- proximately 15% reductions in total energy consumption. These results demonstrate that jointly designed, minimally intrusive control strategies can simultaneously improve reliability and sustain- ability in urban bus operations.

ABSTRACT. Urban service robots in dense urban centers require reliable onboard perception and a trustworthy position claim within connected mobile infrastructure. This work presents a Trusted Position framework integrating a validated lightweight dynamic obstacle detection (DOD) pipeline with 5G connectedness characterization using ns-3/5G-LENA multi-cell simulations under realistic SUMO-based mobility. Quantum Position Verification (QPV) is incorporated as an abstract timing-constrained positional authentication layer. The DOD pipeline sustains 10-15 Hz operation with high static filtering accuracy. Network simulations reveal a throughput-timeliness trade-off: TDMA (Time-Division Multiple Access) schedulers maintain sub-10 ms delivery, whereas OFDMA (Orthogonal Frequency-Division Multiple Access) schedulers increase delays to 85-1505 ms, exceeding freshness budgets for local perception coherence. The results define three trust levels: TDMA trusted, OFDMA MR (Maximum Rate) conditionally trusted, and OFDMA PF (Proportional Fair)/RR (Round Robin) untrusted. Under the studied conditions, the proposed trusted positioning is achievable with TDMA within the proposed timing-bound model.

ABSTRACT. This paper investigates a functional scenario sampling methodology for the validation of Connected and Automated Vehicle (CAV) systems within a scenario-based testing framework. The approach relies on data-driven labelling of critical scenarios and compares two input sources: simulated scenarios generated by a traffic simulator and predicted scenarios derived from machine learning models trained on field data. A four-phase evaluation framework is proposed, including scenario emulation, verification against field data, scenario reduction, and impact assessment on an adaptive cruise control (ACC) use case. The comparison highlights the sensitivity of the scenario selection process to input data, revealing divergences in extrapolated scenarios and potential biases. Results show that both approaches are consistent in observed conditions but differ beyond available data, leading to variations in selected scenarios. This work highlights the importance of input data quality for robust and reliable CAV validation.

ABSTRACT. Motion prediction is essential for safe and reliable decision-making in autonomous driving and robotic systems. Although severe performance degradation is usually expected when a model is deployed to a new environment with different road structures and traffic patterns, collecting onboard driving logs and annotating High-Definition (HD) maps for every new city takes tremendous effort and time. Therefore, this work investigates the potential of using drone-based traffic monitoring as an efficient and scalable data source for transferring motion forecasting models to unseen environments. With occlusion-free drone observations, any traffic participant can be selected as the ego vehicle for training motion forecasting models, yielding high-quality data at much larger volume and with a more diverse distribution. We introduce Songdo Drive, a motion forecasting dataset built from 137 hours of drone traffic monitoring over 20 complex intersections in Songdo, South Korea. Experiments show that a zero-shot model trained on an existing autonomous-driving benchmark degrades significantly when evaluated on Songdo Drive, yet the performance can be substantially improved by leveraging the advantage of drone-based trajectory data.

ABSTRACT. Autonomous shuttles are increasingly deployed for first-last mile connectivity, yet economic assessments assume time valuations transfer directly from isolated to integrated multimodal contexts. We challenge this using a dual-experiment stated choice design examining how trip context shapes autonomous shuttle perception: travelers may value shuttles differently when they serve as access modes versus standalone alternatives. Drawing on 600 respondents across Bayesian D-efficient designs, we test whether favorable AV time valuations documented in isolated settings persist when shuttles function as access links within multimodal networks. Our findings directly address a critical contradiction: while monomodal studies show VTTS reductions for autonomous driving, recent evidence paradoxically finds AVs as last-mile access modes valued worse than manual cars. By revealing how integration within multimodal chains fundamentally alters temporal perception, we provide essential valuation parameters for cost-benefit analyses and expose limits to transferring behavioral insights across trip configurations, critical for deploying autonomous shuttles in complex transport networks.

ABSTRACT. This paper addresses the Service Network Design (SND) problem faced by a Logistics Service Provider (LSP), integrating tactical planning with dynamic operational decisions. The proposed approach considers a complex operational setting, including uncertain travel times and demand, dynamic replanning decisions, and interactions between the LSP and carriers. An agent-based simulation model is developed to capture these complexities and is coupled with a metaheuristic to solve the SND. To overcome the high computational cost of simulation, a Graph Neural Network (GNN) is introduced as a surrogate model. The GNN, trained with simulation-generated data, leverages the network structure to predict both global performance (profit) and node-level indicators, which are used to guide the search process. Numerical experiments show that the proposed framework produces high-quality solutions with scalable computational effort, particularly for large-scale instances. The results highlight the potential of integrating optimization and machine learning to address complex freight transport planning.

ABSTRACT. Predictive generalization is a central challenge in transport choice modeling, especially when models are applied to new settings. One important source of this challenge is out-of-distribution (OOD) prediction, which can arise when data for the same decision problem are collected from different experimental settings. Such settings often produce heterogeneous datasets, including modality-rich and modality-lean data, that provide complementary but incomplete views of behavior. To leverage information from these datasets, we propose a general framework that extends PSIS-LOO stacking to combine predictive distributions from models trained on heterogeneous datasets. Using two stated choice datasets, we find that PSIS-LOO stacking performs best overall. The results further show that the benefit of model fusion depends on the target domain, and that stacking improves performance mainly by correcting severe prediction errors in OOD cases.

ABSTRACT. Urban freight transport contributes substantially to congestion, pollution, and inefficiency in cities. To mitigate these adverse effects, policymakers are increasingly exploring incentive based mechanisms that promote sustainable transportation modes, such as inland waterways and rail-based scheduled services. The effectiveness of such policies critically depends on the strategic interactions among multiple decision making agencies. This study develops a multi agency game theoretic bilevel framework for sustainable city logistics involving transportation authorities, logistics service providers (LSPs), and end customers. The authority determines tax and subsidy policies under a budget constraint. LSPs respond by optimizing mode choice and pricing strategies, while customers select services through a utility maximizing logit model. A case study of Amsterdam compares three feasible implementation strategies and evaluates their effectiveness in improving system efficiency, promoting sustainable mode adoption, and balancing policy feasibility with economic performance.

ABSTRACT. This study introduces the modular share-a-ride problem (MSARP). In this problem, modular vehicles (MVs), individually dedicated to serving either passengers or parcels, can couple and decouple platoons at service nodes, thereby enabling combined passenger and freight transport. The objective is to determine the optimal routing, fleet allocation, and (de)coupling decisions to maximize total profit. The problem is firstly formulated as a mixed-integer linear programming (MILP) model. To solve medium- and large-scale instances efficiently, a Q-learning based adaptive large neighborhood search (QALNS) algorithm is then proposed, which incorporates Q-learning algorithm and novel operators. These optimization components jointly achieve high-quality platoons and solutions. Finally, through extensive numerical experiments, the efficiency of the proposed algorithm are validated. The economic advantage of combined passenger and freight transport is further demonstrated.

ABSTRACT. Every population synthesis method - whether IPF, a Bayesian network, or a deep generative model - produces a fractional distribution over demographic cells. Before agents can be placed in a simulation, those fractions must become whole individuals. This conversion step, which we call integerization, has received little principled attention. We treat it as an inference problem: find the integer table that is closest to the fractional output in Kullback--Leibler (KL) divergence, subject to known demographic margins. We prove that the optimal solution always assigns each cell either its floor or ceiling value. This reduces the problem to a simple two-step algorithm: a second IPF pass followed by deterministic largest-remainder rounding. Applied to 98 Danish municipalities with 635,000 joint cells, the method places 98.4% of population mass in the correct cells, against 92.1% for multinomial sampling. Results are fully reproducible and controlling margins are satisfied exactly.

ABSTRACT. Multi-region bathtub models are widely applied to urban networks with unevenly distributed congestion. They require explicit modeling of regional path choice and flow exchanges. Regional paths are commonly defined as sets of routes that share the same sequence of traversed regions, and their distances can vary depending on traffic conditions. However, existing studies lack a systematic way to capture how regional traffic states affect path distances. This paper extends the multi-region bathtub model by incorporating dynamic path distance approximations for paths in each region. The extended model is evaluated on a partitioned road network of the city of Delft, the Netherlands. Bathtub model simulations with different path distance assumptions are benchmarked against a macroscopic traffic simulation. Results indicate that dynamic path distance approximation improves the accuracy of regional accumulation and speed in regions with large variations in path distance. In regions with limited path distance variation, static distances remain sufficient.

ABSTRACT. Traditional transportation policy evaluations typically focus on the final state of User Equilibrium (UE), assuming it is unique and independent of the implementation process. However, in complex multimodal networks, this uniqueness is often undermined by cross-modal dynamics. This study provides a novel framework to model modal equilibrium by reformulating the problem through a joint probability distribution of trip-length tuples. By shifting the focus from individual origin-destination pairs to trip-length distributions, the framework significantly reduces computational complexity while maintaining spatial accuracy via a trip-based Macroscopic Fundamental Diagram (MFD).

We demonstrate the utility of this approach on the standard problem of designing and implementing pricing zones in dense urban areas. Taking Lyon, France, as a case study, we simulate the phased rollout of congestion pricing to observe the system's evolution. Our findings reveal that the final equilibrium state is inherently path-dependent: the specific sequence of implementation phases fundamentally alters the eventual modal split and network utilization levels. This research contributes both a computationally efficient methodology for multimodal modeling and empirical evidence that the sequence of policy interventions are as critical to urban planning as the final design itself.

ABSTRACT. This study develops a composite suitability framework integrating demand density and flow imbalance to identify areas in the Barcelona Metropolitan Area (AMB) with potential for modal shift from motorized travel to the Bicing bike-sharing system. Motorized origin–destination (OD) flows from traffic analysis zones (TAZ) are redistributed to a uniform rhombus grid and converted into bike-feasible demand using cycling travel times on the road network and a probabilistic bicycle-use model calibrated from observed bike-sharing trip durations. Based on this, demand density and flow imbalance are calculated for each rhombus and combined into a composite suitability score used to classify rhombuses into four categories: Priority, Serve but costly, Stable but weak, and Avoid. Comparison with the current Bicing service area shows strong overlap with Priority areas while revealing substantial high-suitability areas that remain unserved. The framework provides a transparent, scalable approach to identifying strategic bike-sharing expansion or reduction areas.

ABSTRACT. Extreme heat events have been growing in intensity and frequency across the world, posing growing risks to public health, daily mobility, and community resilience. While behavioral adaptation is a critical coping mechanism during these events, the understanding of how prior experience shapes future adaptation choices is limited. This study addresses this by examining how the severity of past heat events influences anticipated future activity-travel choices. Using a nationwide survey data from the U.S., two multivariate econometric model systems are estimated – one for mobility choices and one for activity participation. Results reveal that perceived severity of prior extreme heat events directly affects anticipated behavioral adjustments, with individuals learning from experience to reduce heat exposure through decreased transit use, increased car use, staying home, and greater participation in indoor activities. It also found that vulnerable groups perceive heat impacts more intensely yet exhibit limited adaptation capacity, underscoring the need for targeted interventions.

ABSTRACT. This paper formulates daily activity scheduling—decisions on when, where, and for what purpose to perform activities and travel—as a Perturbed Utility Route Choice (PURC) problem on a time–space network. Unlike conventional discrete choice models, the proposed framework represents behavior directly as link flows, corresponding to the number of individuals at each location and time. In this framework, optimal flows are sparse, allowing the model to endogenously restrict the effective consideration set without ex ante enumeration while capturing flexible substitution and complementarity patterns across schedules. To address the large size of time–space networks, we propose a column-generation-based estimation algorithm that iteratively expands partial networks. We further introduce a group-based perturbation structure to capture activity-specific correlations beyond those implied by network topology. An empirical application using the Danish National Travel Survey demonstrates that the model can be estimated on realistic networks and reproduces key activity–travel patterns.

ABSTRACT. Understanding how individuals make travel decisions is central to transportation research. Discrete choice models provide a rigorous framework to infer preferences from observed behavior and to anticipate responses to policies. However, their practical application fundamentally relies on the ability to enumerate or sample from a predefined choice set. This assumption breaks down in many important contexts, such as route choice on large-scale networks or activity scheduling, where the number of feasible alternatives is combinatorial.

This paper addresses this limitation by introducing estimation methods that do not require explicit enumeration of the choice set. Instead of treating the choice set as a fixed input, we embed alternative generation directly within the estimation process.

We develop two complementary approaches. The first relies on adaptive sampling within a Bayesian framework, using only binary comparisons to recover the full model. The second leverages Bayesian optimization for likelihood-free inference (BOLFI), replacing likelihood evaluation with a learned surrogate of the discrepancy between observed and simulated behavior. Both approaches fundamentally bypass the need to enumerate or approximate the full choice set.

Using controlled synthetic experiments with known ground truth, we show that these methods recover model parameters with the same accuracy as full choice set estimation, while achieving orders-of-magnitude computational gains. Crucially, their performance scales independently of the size of the choice set. These results demonstrate that discrete choice models can be estimated reliably without enumerating alternatives, substantially expanding the range of problems that can be addressed in practice. In particular, they enable the estimation of behaviorally rich models in settings previously considered intractable, including network-wide route choice and combinatorial activity scheduling.

ABSTRACT. Simulation-based transport planning often relies on transferring mode choice models between regions, particularly when studying new or emerging transport modes such as autonomous vehicles (AVs), for which observed behavioral data are scarce. The validity of such transfers, however, remains uncertain. This study empirically assesses the transferability of discrete mode choice models including AV alternatives across Flanders, Cantabria, the Netherlands, Athens, Kansas City, Seattle, and New York City. Comparable multinomial logit models were estimated from stated preference datasets, and cross-regional equivalence was evaluated using likelihood ratio tests. Results show that behavioral parameter equivalence across regions is generally rejected, with substantial heterogeneity observed for most transcontinental comparisons and among U.S. metropolitan areas. Evidence consistent with behavioral equivalence is found primarily within Europe, particularly between Belgium and the Netherlands. These findings underscore the need for caution when applying transferred mode choice models in simulation and policy analysis.

ABSTRACT. The widespread adoption of autonomous vehicles (AVs) is expected to fundamentally reshape how individuals perceive and value their travel time. However, the psychological constructs that interpret this relationship have not been sufficiently studied. This study examines the influence of attitudinal factors, specifically non-shared autonomous vehicle (NSAV) acceptance and technology trust, on travel time preferences and the Value of Time (VOT) for AVs. Using stated preference (SP) survey data from Oxfordshire and the West Midlands, England, we developed a hybrid choice model (HCM). This model analyzes user choices among four transportation modes: private vehicle (PV), public transit (PT), private autonomous vehicle (PAV), and shared autonomous vehicle (SAV) while including latent psychological constructs derived from attitudinal indicators. The modeling results show that NSAV acceptance is a strong and significant predictor of both PAV and SAV adoption, while trust indicates a significant negative effect on SAV choice, suggesting that concerns about privacy and shared riding reduce willingness to use shared automated services. VOT estimates exhibit meaningful heterogeneity across income, age, and gender, consistent with the model's socio-demographic interaction structure. In terms of AV features, safety perception emerged as a marginally significant positive attribute, while connectivity and dynamic routing features were found to be statistically insignificant. These findings highlight the importance of including psychological factors alongside conventional travel characteristics when modeling the adoption of AVs and planning for such services, providing practical implications for service providers and policymakers in the design of future autonomous transportation.

ABSTRACT. Germany’s 9-Euro-Ticket, implemented from June to August 2022, attracted exceptional international attention because of both the magnitude of the fare discount and the scale of its implementation. It also sparked intense debate over whether the policy mainly generated additional public transport demand or shifted travel away from private cars. This paper estimates the causal impact of the policy on travel behaviour using a longitudinal smartphone-based GPS panel covering May to August 2022. GPS data offer rich, high-frequency, user-level information on trips, mode use, and distance travelled, making them especially valuable for policy evaluation. At the same time, they pose important challenges for causal inference: some users stop contributing data, reporting may be intermittent, and certain outcomes may be observed selectively in ways that are related to travel behaviour itself. If left unaddressed, these features of the data-generating and measurement process can bias estimated policy effects and obscure their dynamics. To address this problem, we develop an estimation framework that treats panel dropout and selective outcome observation as explicit features of the measurement system. The approach constructs stabilised inverse-probability weights on a balanced user-time panel and combines them with a sharp regression discontinuity in time design. Applied to the 9-Euro-Ticket, the results show that conclusions depend strongly on how travel demand is measured. Public transport mode share rises substantially, but the frequency of car use does not fall in any trip-purpose or distance-band category. Although public transport distance increases while car kilometres decline, a temporal decomposition shows that these changes do not occur at the same time, providing little evidence of direct mode shift. More broadly, the paper contributes a practical weighting and diagnostics pipeline that can be embedded in other causal designs using smartphone travel data, helping support more credible policy evaluation.

ABSTRACT. Bus and rail are prominent modes of public transport in Europe for intercity travel. Their modal networks differ substantially in speed, structure and geographic reach, yet are rarely compared systematically at the continental scale. This study constructs a harmonised, service-level intercity network for 134 cities across 30 European countries using timetable data from major intercity rail and coach operators. Generalised travel time and population-weighted accessibility are computed across city pairs, modes and transfer levels. Results show that rail is considerably faster but spatially concentrated in western and central Europe, while bus provides broader coverage across peripheral regions, including the Balkans, Baltics and Iberian cross-border corridors. Rail accessibility saturates around 15 hours of travel time while bus saturates much later at 40 hours. Intermodal transfers generally fare between bus-only and rail-only connections but outperform them in accessibility for more than two transfers, highlighting the critical role of cross-modal coordination in pan-European connectivity.

ABSTRACT. Demand-responsive transport (DRT) is typically routed by solving Dynamic Vehicle Routing Problems (DVRPs), where individual vehicle trajectories are adjusted on incoming requests. This limits demand consolidation and thus efficiency. On the other hand, Conventional Public Transport (CPT) bus systems are based on a network of lines and users find their routes on it, which provides high demand consolidation. However, such a network is built offline and cannot adapt to the demand. We propose a public transport management strategy that reconciles efficiency and adaptivity by dynamically designing a structured network of lines via a receding-horizon optimization approach. Using real-world trip requests, we show that we nearly double the fraction of served requests compared to DVRP-based routing, and we serve more requests than CPT with lower user trip times.

ABSTRACT. Managing cordon queues and spillback is one of the major challenges in perimeter control (PC). To address this issue, this study proposes a two-stage framework for signal optimization at perimeter intersections. In Stage I, the concept of network buffer capacity is introduced, together with a calibration method. Based on this concept, the optimal gated inflows are determined through a spillback-aware multi-objective model to jointly account for network- and cordon-level control objectives. The weight coefficients in the model are adaptively adjusted according to the real-time traffic states. In Stage II, the green durations and phase structures of each perimeter intersection are simultaneously optimized using a group-based signal control approach. Experiments show that, compared with traditional feedback PC strategies, the proposed framework significantly mitigates cordon spillback while preventing network gridlock. The results also highlight the necessity of jointly optimizing phase structures and green durations under time-varying traffic conditions.

ABSTRACT. Reliable and precise accident data are a crucial prerequisite for mitigating accident-related costs and improving the safety of the road traffic network. However, in practice, reported accident data often suffer from temporal and spatial uncertainties, creating a gap between when and where accidents are recorded and when and where their effects are observable in traffic flow and congestion. To address this challenge, we propose a Bayesian multi-source data fusion framework that reconstructs latent accident attributes by integrating heterogeneous evidence while explicitly quantifying uncertainty. Key contributions of this work are: (i) a latent-variable formulation under Bayesian inference that frames accident records as uncertain observations rather than deterministic inputs, (ii) a scalable multi-source data fusion approach that integrates matrix signaling information data and traffic measurements as complementary probabilistic evidence, and (iii) the resulting uncertainty-aware accident dataset can be directly used in data-driven models, such as Bayesian networks, leading to more reliable robust inference in safety and accident management applications.

ABSTRACT. Urban traffic monitoring requires accurate network-wide estimates even though only a limited portion of the road network can be observed in real time. At the same time, the growing deployment of Unmanned Aerial Vehicles (UAVs) in everyday applications is creating new opportunities for high-resolution, real-time traffic monitoring. Compared with traditional fixed loop detectors, fleets of UAVs acting as mobile sensing networks offer substantially greater flexibility, improved accuracy, and broader spatial coverage. In this paper, we propose a Gaussian-process-based framework for real-time estimation and short-term prediction of urban traffic variables from partial drone observations. The area of interest is represented as a weighted graph, and the proposed covariance structure combines a graph Matern kernel, which captures the topology of the road network, with a temporal kernel that models the evolution of traffic over time. In addition to providing network-wide estimates and predictions, the proposed framework also yields a principled quantification of predictive uncertainty at unobserved locations. The model is trained and evaluated using micro-simulations of Barcelona’s urban road network under multiple demand scenarios, demonstrating the effectiveness of Gaussian processes for traffic parameter estimation and prediction.

ABSTRACT. Bicycle route models often face a trade-off between computational cost and the inclusion of the many variables affecting route choice. We propose a scalable multicriteria routing model designed for large-scale applications like traffic assignment and network design. By extending the weighted shortest path approach, the framework leverages trajectory data alongside optimization and machine learning to estimate edge weights. Improvements are observed across all metrics (up to 13%) compared to the distance-based shortest path, with the neural network performing best among the learning models. These cost estimation models are network-agnostic; thus, they can adapt to altered edge attributes and are transferable to unseen regions as they rely on open data. Moreover, a perturbed variant is presented to enhance aggregate bicycle count accuracy. By avoiding costly choice-set construction, this framework provides a computationally efficient, data-driven solution for bicycle route generation.

ABSTRACT. The growing adoption of electric bike-sharing systems introduces new challenges in managing bike availability due to state-of-charge (SOC) dynamics. Existing models typically treat bikes as homogeneous and do not capture the interaction between SOC evolution and system performance. This study addresses charger placement through a unified framework that integrates SOC dynamics, ridership estimation, and infrastructure deployment within a two-dimensional Markovian system. Steady-state probabilities are derived to estimate expected ridership, which serves as the objective in a station-level location optimization problem. A heuristic solution method combined with a single-pooling approximation ensures scalability for large networks. Results show that the proposed approach achieves near-optimal performance, within 0.5–1.5\% of the global optimum. Increasing charger deployment improves availability and accessibility, although with diminishing returns. Case studies in Pittsburgh, Vancouver, and San Francisco demonstrate that SOC-aware charger placement significantly reduces low-SOC occurrences and adapts to different demand patterns.

ABSTRACT. Pedestrian safety at midblock crossings is a critical concern in mixed traffic environments where autonomous vehicles (AVs) and human-driven vehicles (HDVs) share the road. Pedestrians often infer intent from vehicle motion in AV encounters, making them vulnerable to small shifts in conflict margins. This study investigates whether virtual reality (VR) crossing sessions separate into distinct interaction risk profiles and whether AV-only sessions shift profile prevalence compared to HDV-only sessions. Using large-scale immersive VR experiments from Toronto, Canada, and Newcastle, England, we compute surrogate safety measures (SSMs) and apply latent profile analysis (LPA) to identify distinct pedestrian crossing stances, ranging from risk-accepting to highly cautious. Key findings show that Newcastle exhibits a higher prevalence of high-urgency risk profiles in AV-only sessions, indicating that AVs contribute to higher-risk encounters. In contrast, Toronto shows no significant difference between AV-only and HDV-only sessions, suggesting that contextual factors influence the impact of AVs on pedestrian safety.

ABSTRACT. Egocentric vision captures pedestrian visual perception and behavior from a first-person perspective, offering fine-grained insights crucial for pedestrian intention modeling. However, its application in traffic safety remains largely unexplored. We reformulate pedestrian crossing intention prediction as a Visual Question Answering (VQA) task, leveraging the pre-trained knowledge and reasoning abilities of Vision-Language Models (VLMs). We first benchmark two state-of-the-art VLMs in a zero-shot setting, finding that they achieve moderate gains over random guessing in accuracy but exhibit limited higher-level reasoning (e.g., vehicle dynamics). Built upon this observation, we employ parameter-efficient fine-tuning to adapt VLMs to the task. Our results show that the adapted models significantly outperform zero-shot approaches and achieve a 9\% relative improvement in accuracy over a specialized transformer-based baseline. Finally, we demonstrate that integrating dynamic eye-gaze signals and personal attributes further boosts predictive accuracy, establishing a new state-of-the-art for egocentric intent decoding.

ABSTRACT. Two- and three-wheel vehicle users are generally considered a homogeneous group in traffic safety research. However, evidence supports that rider behaviors vary largely based on situational factors such as interactions with vehicles, other riders, and the environment. Identifying latent behavioral subgroups can therefore provide a more meaningful basis for understanding behavioral patterns among riders, as well as developing targeted safety interventions. This study aims to uncover behavioral patterns among two- and three-wheel vehicle users within Copenhagen’s bicycle lanes. Observational video data collected in 2025 at four locations were analyzed using Latent Class Analysis to identify distinct behavioral groups based on vehicle, environmental, and sociodemographic factors. Three resulting classes were then described based on the profile of their observed characteristics. Findings from this research provide insights that can be used to inform targeted safety and policy interventions as a means of addressing subgroup specific behaviors among two- and three-wheel vehicle users.

ABSTRACT. High-quality and complete representations of cycling infrastructure are essential for transport research, planning, and publicly funded projects. In many countries, OpenStreetMap (OSM) serves as a primary open data source for modelling and decision support in this domain. At the same time, limitations of OSM data completeness – especially for cycling infrastructure – are frequently reported. Detecting infrastructure that is entirely missing from OSM remains particularly challenging, as such gaps cannot be identified through internal data consistency checks alone, while comprehensive authoritative reference datasets are often unavailable. This paper presents a methodology for systematically detecting such gaps by combining automated object detections from crowdsourced street-level imagery with spatial analysis of the existing OSM road network. Traffic signs and road markings serve as indicators of cycling infrastructure not yet represented in OSM. Candidate locations are filtered through spatial, temporal, and contextual criteria and translated into community validation tasks via MapRoulette. Applied across Germany, the approach generated over 1,200 validation tasks with false-positive rates below 20 % among completed tasks. Through the resulting campaigns, more than 500 km of previously unmapped cycling infrastructure were added to OpenStreetMap.

ABSTRACT. This study proposes an innovative methodology for evaluating sidewalk efficiency through the assessment of the Pedestrian Level of Service (PLOS) as experienced by pedestrians. The approach is structured around four key pillars: infrastructure accessibility, convenience of movement, safety and security. It also aims to capture pedestrians’ preferences regarding their walking experience in order to estimate weighting coefficients and develop an index for quantifying pedestrian service levels. A questionnaire survey was conducted to examine factors influencing pedestrian experience, considering built environment characteristics, infrastructure and traffic conditions and user demographics. The findings indicate that gen-der affects the factors that pedestrian prioritise, with women placing greater importance on security as well as convenience. Regarding age, convenience becomes increasingly im-portant as pedestrians grow older. Overall, the results contribute to the development of a standardized framework for assessing sidewalk performance and identifying critical factors affecting pedestrian experience.

ABSTRACT. Open science initiatives have strengthened scientific integrity across many fields, yet their adoption in transportation research remains under-investigated. Key features of open science---data and code availability---are difficult to extract at scale due to the complexity of full-text publications. We introduce a scalable, LLM-based pipeline to measure data and code availability and validate its performance through inter-rater agreement analysis against a manually curated dataset of 96 papers. Applied to 10,724 research articles published in Transportation Research journals (2019--2024), we find that only 5% of quantitative papers share a code repository and 4% share a data repository, with patterns varying across journals, topics, and geographic regions. No significant link is found between open science practices and traditional academic incentives, suggesting that structural interventions from publishers and funding agencies are needed.