Urban Currents

TrajGPT-R: Generating urban mobility trajectory with reinforcement Learning-Enhanced generative Pre-trained transformer

The researchers developed TrajGPT-R, a transformer-based framework that generates urban mobility trajectories by framing the task as an offline reinforcement learning problem. The model reduces vocabulary space during tokenization and uses Inverse Reinforcement Learning to capture trajectory-wise reward signals from historical data, which are then used to fine-tune the pre-trained model. Evaluations across multiple datasets demonstrate that this approach outperforms existing generative models in trajectory reliability and diversity.

Why it matters — It provides a robust methodology for simulating realistic, privacy-preserving urban mobility data, overcoming traditional reinforcement learning limitations like sparse rewards and long-term credit assignment in autoregressive generation.

Transportation Research Part C Emerging Technologies · doi · code

Imagining economic development, enabling sprawl: Planning for logistics in suburban regions

This study analyzes how local governments facilitate industrial expansion at the rural-suburban fringe through a qualitative case study of a southeastern suburb in British Columbia's Lower Mainland, Canada. The researchers examined municipal planning documents, newspaper archives, real estate publications, Census data, and conducted interviews with local planning officials to trace how municipal efforts accommodate large-scale logistics facilities and warehouses.

Why it matters — It challenges the conventional urban planning literature that primarily attributes suburban edge expansion to residential growth, demonstrating instead how local officials' economic imaginaries and global logistics demands drive industrial sprawl.

Caveat: The findings are based on a single suburban case study in Canada, which may limit direct generalizability to regions with different municipal governance structures or land-use regulations.

Cities · doi

Urban green credits: Capturing ecosystem service value in housing markets for redistributive urban planning

This study introduces Urban Green Credits (UGC), a planning framework that connects ecological performance to fiscal policy using a parcel-level Index of Ecosystem Services (IES) that measures carbon sequestration, thermal regulation, ecological connectivity, and green space accessibility. Applying hedonic, spatial econometric, instrumental variable, and difference-in-differences models to 5,230 residential transactions in Porto, Portugal, the researchers quantified how ecosystem services capitalize into housing prices and simulated the revenue potential of capturing these values.

Why it matters — It establishes a concrete mechanism to capture the privately realized real estate value generated by public green infrastructure, offering a novel pathway to self-finance urban greening and redistribute resources to ecologically underserved neighborhoods.

Caveat: The proposed framework is conceptual and requires local calibration before it can be applied to other urban land markets.

Cities · pdf · doi

Revisiting the effects of the built environment on VMT in California using quantile regression

Using data from the 2010–2012 California Household Travel Survey, this study applies quantile regression models to analyze how built environment features affect daily personal-level vehicle miles traveled (VMT). It finds that a one-unit increase in population density (representing 1,000 people per square meter in 20-minute driving accessibility) correlates with a 10% to 12.5% VMT reduction for high-travel individuals (above the 0.60 quantile), which is four times larger than the 2.5% reduction seen at the 0.20 quantile. Additionally, this density increase is associated with up to an 18% reduction in driving-others VMT at the 0.80 quantile and up to a 12.5% reduction in passenger VMT at the 0.85 quantile.

Why it matters — The study demonstrates that urban density and land-use policies do not affect driving behavior uniformly, showing that built environment interventions are significantly more effective at reducing mileage among heavy drivers than light drivers.

Caveat: The analysis relies on travel survey data from 2010–2012, which may not fully reflect post-pandemic travel patterns and spatial dynamics in California.

Research in Transportation Economics · doi

Road rules, infrastructure and risk − how law-abiding are e-scooter riders?

This study analyzed over 600,000 traffic camera observations of e-scooter riders, cyclists, and pedestrians across eight sites in Brisbane, Australia, over a two-year period. Using machine learning for initial processing and logistic regression for modeling, the research examined how helmet use and speed compliance relate to infrastructure type, urban density, and vehicle ownership (shared versus private). The analysis found that speed compliance is highest on separated cycle tracks, and that riders wearing helmets—especially full-face helmets—travel at higher speeds.

Why it matters — It provides empirical evidence of risk compensation behavior among e-scooter riders, demonstrating that riders adjust their speed and helmet use based on infrastructure type. It also establishes that building physically separated cycle tracks directly encourages higher rates of both speed and helmet compliance.

Caveat: The findings are based on observations from eight specific camera locations within a single city, which may limit generalizability to other urban regulatory and infrastructural contexts.

Travel Behaviour and Society · pdf · doi

Beyond solo UAM flights: Service preferences incorporating price sensitivity and user behaviors

This study analyzed user preferences and price sensitivity for shared versus solo Urban Air Mobility (UAM) services using a stated preference survey. The researchers developed a hybrid choice model that integrates demographic interactions, environmental awareness, and price sensitivity to evaluate trade-offs between travel time, cost, waiting time, and passenger capacity.

Why it matters — It establishes that environmental awareness and demographic-specific price sensitivity are key mediating factors in UAM adoption, demonstrating that shared flight models can lower per-passenger costs through higher vehicle utilization while appealing to environmentally conscious travelers.

Caveat: The findings are based on stated preference survey data rather than observed travel behavior in an active, real-world UAM market.

Transportation Research Part A Policy and Practice · pdf · doi

How does parking regulation policy affect e-scooter sharing systems: A data-driven approach and analysis in Nordic cities

This study evaluates the impact of e-scooter parking regulations across three Swedish cities using a framework that combines regression discontinuity design and difference of probability. The analysis reveals that the policies decreased overall e-scooter demand, particularly in city centers, while simultaneously increasing average travel speeds and reducing both trip durations and distances.

Why it matters — It quantifies how regulatory constraints shift micro-mobility behavior, demonstrating that while policies may suppress overall ridership, they can increase per-trip efficiency and reduce greenhouse gas emissions per trip by encouraging the substitution of less sustainable transport modes.

Caveat: The findings are derived from three specific Nordic cities and may not directly generalize to urban areas with different transit infrastructures, regulatory baselines, or climate conditions.

Transport Policy · doi

Rent gap thinking beyond gentrification: rent gap deflation in Havana, Cuba

This study analyzes the housing market in Havana, Cuba, following the 2011 reforms that legalized home sales and tourism-based property uses. It documents a process of rent gap deflation, where potential ground rents are shrinking rather than expanding due to economic crisis, geopolitical isolation, declining tourism, and mass emigration.

Why it matters — The paper challenges the assumption that rent gaps inevitably lead to gentrification, demonstrating how restricted property rights, state land ownership, and macroeconomic volatility can instead produce arrested gentrification and capital withdrawal.

Caveat: The findings are based on a qualitative conceptual analysis of Havana's unique post-socialist transition, which may limit direct generalizability to more conventional market economies.

City · pdf · doi

A two-stage hybrid-computing framework for generating metro-style tourist maps: integrating a fine-tuned LLM with geospatial optimization

A two-stage hybrid-computing framework was developed to generate customized, metro-style tourist route maps. The system first uses a tourism-specific large language model, fine-tuned on attraction descriptions and route planning Q&A datasets, to output structured travel recommendations, and then applies Voronoi-diagram-based and grid-based optimization algorithms to render the schematic maps. The framework's performance was validated using BLEU-4 metrics and subjective surveys completed by over 300 participants.

Why it matters — It automates the creation of highly legible, schematic tourist maps tailored to individual user demands, a capability missing from current digital mapping platforms that rely on standard geographic layouts.

International Journal of Geographical Information Systems · doi

Multisource data-driven diagnostic assessment of Hangzhou’s cycling system: A zonal policy framework for smart low-carbon urban mobility

The study establishes a diagnostic framework to evaluate Hangzhou's cycling system using 12 indicators across convenience, vitality, and safety dimensions. It integrates diverse geospatial datasets, including OpenStreetMap bicycle lanes, NDVI for tree-shaded pathways, POI data for bike-sharing stations, and Gaode real-time traffic volume indices, to analyze performance across four distinct functional zones.

Why it matters — It provides a zone-specific, adaptive governance framework that moves beyond uniform city-wide planning, allowing planners to target infrastructure investments—such as shaded corridors, metro connections, or safety measures—based on localized performance profiles and a documented center-periphery disparity.

Caveat: The diagnostic framework relies heavily on real-time traffic indices and POI data from specific commercial providers, which may limit its direct reproducibility in regions where these proprietary data sources are unavailable.

International Journal of Sustainable Transportation · doi

The relationship between zoning-based urban expansion and vehicle energy consumption: The spatial outcome of green area vs. commercial–industrial–residential area expansion

This study analyzed 11,117 household-year observations from the Household Energy Panel Survey (2018–2022) using a panel hierarchical linear model to measure how different zoning expansion pathways affect vehicle energy consumption. The analysis compared commercial-industrial-residential (CIR) expansion against green area expansion across three pathways: incorporating non-urban land, converting existing zones, and designating previously undesignated areas. The results show that expanding green areas reduces household vehicle energy consumption with an effect size 14.8 times larger than the consumption-increasing effect of CIR expansion.

Why it matters — It quantifies the specific transport energy impacts of different land-use conversion pathways, proving that incorporating non-urban land into green space actively lowers driving energy while converting it to CIR uses increases fuel consumption, especially for apartment dwellers.

Caveat: The study relies on household energy survey data from a single country context, though the abstract does not name the specific nation.

International Journal of Sustainable Transportation · doi

Koopman-enhanced spaciotemporal traffic prediction model for high-order phenomena

The study develops KMD-CTM, a hybrid physical framework that integrates Koopman mode decomposition with the cell transmission model to predict traffic density and flow. The model incorporates speed-dependent headway distributions to capture congested discharge dynamics, hysteresis, and capacity drops while maintaining physical consistency between upstream and downstream traffic.

Why it matters — It provides a computationally efficient traffic prediction method that requires less offline training and hyperparameter tuning than deep learning models, while remaining robust to noisy or missing data and adhering to traffic equilibrium principles.

Transportmetrica B Transport Dynamics · doi

Nonlinear determinants of discretionary activity episodes in non-mandatory tours

The study introduces a Neural-Network-embedded Generalised Ordered Response Probit (NNGORP) model to capture nonlinear effects on discrete activity counts. Using 60,309 tours from the 2019 Shanghai Household Travel Survey, the model analyzes how sociodemographic, household, and built-environment factors influence discretionary activity frequency, revealing threshold responses for variables like age, bicycle ownership, and neighborhood density.

Why it matters — It provides a method to model complex, nonlinear travel behaviors within a coherent probabilistic framework, yielding structurally derived partial dependence patterns that are more behaviorally consistent and generalizable than post-hoc machine learning explanations like SHAP.

Caveat: The empirical application and validation of the model are limited to a single high-density urban context using data from Shanghai.

Transportmetrica A Transport Science · doi

Calming traffic in unsignalized bidirectional streets by introducing one-lane bottlenecks

This study models the performance of first-in-first-out (FIFO) and directional priority (DP) traffic policies at single and multiple one-lane bottlenecks on unsignalized, bidirectional streets. It evaluates expected delay under low demand and system capacity under oversaturated conditions, deriving mathematical formulas for capacity boundaries. The analysis reveals that while FIFO causes less delay than DP under low demand, DP can achieve higher capacity under balanced high demand, and adding extra bottlenecks can unexpectedly increase capacity for longer streets under DP.

Why it matters — It provides analytical formulas that allow traffic planners to mathematically compare the capacity and delay trade-offs of different bottleneck priority rules, revealing counterintuitive capacity dynamics when multiple bottlenecks are chained together.

Caveat: The findings are based on theoretical mathematical modeling of traffic flows rather than empirical field observations or microsimulations.

arXiv · pdf

Proactive Road Safety Intervention in Australia: Predicting Risky Driving Hotspots from Connected Vehicle Data

The study analyzed connected vehicle telemetry data from Greater Sydney, Australia, to map and forecast near-miss driving events across Local Government Areas using g-force thresholds for hard braking, harsh cornering, and harsh acceleration. Eight predictive models were benchmarked, with classical ARIMA achieving the lowest mean absolute error of 162.21, closely followed by LSTM at 163.92, while identifying persistent high-risk zones in Sydney's central business district, Parramatta, and Bankstown.

Why it matters — It demonstrates that proactive road safety monitoring can be achieved using real-time vehicle telemetry before crashes occur, and proves that simpler time-series models can perform as well as deep learning methods when training data is constrained.

Caveat: The predictive modeling and risk mapping are aggregated at the relatively coarse Local Government Area level rather than specific intersections or street segments.

arXiv · pdf

Design of policy digital twins incorporating multi-level agent based modelling

This paper presents a design framework for policy digital twins that integrates multi-level agent-based modelling to capture human behaviors and their impacts on policy outcomes. The framework is demonstrated through a case study of a digital twin designed for a UK city council to assist in planning and delivering its energy transition policy.

Why it matters — It addresses the slow adoption of digital twins in public policy by providing a structured design method that explicitly incorporates multi-level human decision-making, a critical factor often omitted in traditional engineering-focused digital twins.

Caveat: The paper presents the design method and architecture of the digital twin rather than empirical performance metrics or deployment evaluations of the system in active policy-making.

arXiv · pdf

Impacts of social and impact heterogeneity on social-climate outcomes

Using a coupled social-climate model across five global regions, this study measures how regional variations in social traits, temperature shifts, and climate vulnerability alter global warming projections. Incorporating climate impact heterogeneity increases the projected global temperature anomaly by 0.2°C, while adding social heterogeneity raises it by another 0.1°C.

Why it matters — The research demonstrates that ignoring regional disparities underestimates global temperature rises, proving that reducing global social and climate inequalities directly improves physical climate outcomes.

Caveat: The findings are based on a coupled social-climate simulation model rather than empirical historical data.

arXiv · pdf