Welcome!
Designing cycling infrastructure requires balancing the competing needs of diverse user groups, yet designers often struggle to anticipate how different cyclists experience the same street environment. We investigate how persona-based evaluation can support cycling infrastructure design by making experiential conflicts explicit during the design process. Informed by a formative study with 12 domain experts and crowdsourced bikeability assessments from 427 cyclists, we present StreetDesignAI, an interactive system that enables designers to (1) ground evaluation in real street context through imagery and map data, (2) receive parallel feedback from simulated cyclist personas spanning confident to cautious users, and (3) iteratively modify designs while the system surfaces conflicts across perspectives. A within-subjects study with 26 transportation professionals comparing StreetDesignAI against a general-purpose AI chatbot demonstrates that structured multi-perspective feedback significantly Broaden designers’ understanding of various cyclists’ perspectives, ability to identify diverse persona needs, and confidence in translating those needs into design decisions.
Key words: Cycling Infrastructure, Multi-Perspective Design, Persona-based Evaluation, Multi-agent System, Generative AI
Wang, Z., Dai, Y., Lyu, D., Nader, M., Chen, S., Ye, W., Ding, Z., Yan, X. (2026). StreetDesignAI: A Multi-Persona Evaluation System for Inclusive Infrastructure Design. In Proceedings of the 2026 ACM Designing Interactive Systems Conference (DIS ’26). ACM. Download Preprint.
Advancing human-centered infrastructure requires assessment approaches that consider users’ subjective perceptions and preferences. Bikeability assessment is a representative case where user-centric evaluation must incorporate perceived safety and comfort across diverse cycling groups. Existing perception-based bikeability assessments face fundamental challenges for which Multi-modal large language models (MLLMs) offer a promising solution. However, the application of MLLMs in this domain faces another challenge: lack of supervision signal. That is, existing bikeability assessment approaches and datasets only provide scalar ratings, with no labels for factor attribution or reasoning traces, leaving MLLMs without the supervision signal needed to learn accurate rating prediction and robust chain-of-thought reasoning. In this paper, we propose a persona-conditioned framework that closes these supervision gaps and fine-tunes an MLLM end-to-end. The framework consists of: (i) counterfactual street-view augmentation for data scaling and per-feature attribution; (ii) an LLM-normalized factor vocabulary that unifies controlled and free-text survey responses; and (iii) persona-conditioned multi-granularity fine-tuning that jointly predicts ratings, identifies influencing factors, and generates explanations.
Key words: Bikeability; LLM, Generative AI, chain-of-thought, human-centered infrastructure
Dai, Y., Wang, Z., Wang, C., Zhou, K., Qian, Y., Xu, S., Yan, X. Persona-Conditioned and Explainable Bikeability Assessment: A Multi-modal LLM Approach. Download Preprint.
We have developed an automated, low-cost, and generalizable approach using Google Street View images and deep learning techniques to evaluate bus stop amenities. Leveraging the latest YOLOv8 model, transfer learning, and a dynamic prediction algorithm, our approach achieves efficient detection of bus stop amenities (e.g., shelters and benches) with high accuracy and precision. Scalability and transferability tests further suggest that highly accurate feature detection results can be achieved through model fine-tuning on a small sample of local data.
Key words: Computer Vision, transit, Google Street View Image, Bus Stop Amenities
Dai, Y., Liu, L., Wang, K., L., M., Yan, X. Using Deep Learning and Google Street View Images to Assess Bus Stop Amenities. Download Preprint.
As AI is increasingly used in transportation, it is important to understand how transportation professionals, the driving force behind AI Transportation applications, perceive AI’s potential efficiency and equity impacts. We surveyed 354 transportation professionals in the United States and conducted descriptive analysis and latent class cluster analysis based on the collected data.
Key words: AI, transportation, equity, latent class cluster analysis
Qian, Y., Polimetla, T., Sanchez, T., Yan, X. (2024). How do transportation professionals perceive the impacts of AI applications in transportation? A latent class cluster analysis.. Download Preprint.
With increased frequency and intensity due to climate change, wildfires have become a growing global concern. This study proposes a new methodology to analyze human behavior during wildfires by leveraging a large-scale GPS dataset. This methodology includes a home-location inference algorithm and an evacuation-behavior inference algorithm, to systematically identify different groups of wildfire evacuees (i.e., self-evacuee, shadow evacuee, evacuee under warning, and ordered evacuee).
Key words: wildfire, GPS data, evacuation, human behavior
Zhao, X., Xu, Y., Lovreglio, R., Kuligowski, E., Nilsson, D., Cova, T., Wu, A., Yan, X., Cao, Z. Estimating wildfire evacuation decision and departure timing using large-scale GPS data. [Download Preprint]. Transportation research part D: transport and environment, 107, 103277.
We develop an analytical framework to examine how dockless e-scooter and station-based bikesharing differ regarding a set of equity-related outcomes (i.e., availability, accessibility, usage, and idle time) across neighborhoods in different socioeconomic categories. An analysis of idle time is made possible by the availability of the GBFS data, a new source of open big data. The analysis of idle time can shed light on if improving spatial access to shared micromobility vechicles in low-income communities can effectively promote micromobility use in these areas.
Key words: E-scooter, bikeshare, micromobility equity
Su, L., Yan, X., Zhao, X. Micromobility equity: A comparison of shared e-scooters and station-based bikeshare in Washington DC. [Download Preprint]. Transport policy, 145, 25-36.
We a crowdsourcing platform [watch a 3-min demo that have two main functions]:
1) It allows residents to report micromobility-related safety incidents (i.e., crashes and near-misses).
2) It allows city transportation staff to conduct analytics regarding e-scooter parking and usage patterns.
In Feburary 2019, I developed a Python program that automatically download the General Bikeshare Feed Specification (GBFS) data for multiple U.S. cities. In this study, we developed algorithms to infer micromobility trip origins and destinations based on the GBFS data.
Key words: micromobility, GBFS data, e-scooter
Xu, Y., Yan, X., Sisiopiku, V., Merlin, L., Xing, F., Zhao, X. Micromobility trip origin and destination inference using General Bikeshare Feed Specification (GBFS) data. [Download Preprint].
In this paper, we developed a random forest model to examine which factors are key predictors of ridesplitting adoption rate (i.e., the proportion of ridesourcing trips with ridesharing authorization) and to explore their nonlinear relationships. The random forest model resulted in additional insights that were not obtainable from the commonly applied linear regression model.
Key words: random forest, machine learning, travel demand prediction
Xu, Y., Yan, X., Liu, X., Zhao, X. (2021). Identifying key factors associated with ride-splitting rate and modeling their nonlinear relationships. [Download Preprint]. Transportation Research Part A: Policy and Practice, 144, 170-188. https://doi.org/10.10416/j.tra.2020.12.005
The increasing popularity of machine learning in transportation research raises questions regarding its advantages and disadvantages compared to conventional logit-family models used for travel behavioral analysis. We provide a comprehensive comparison between logit models and machine learning by examining the key differences in model development, evaluation, and behavioral interpretation in mode-choice modeling. We find that There appears to be a tradeoff between predictive accuracy and behavioral soundness when choosing between machine learning and logit models in mode-choice modeling.
Key words: machine learning, logit model, travel mode choice
Zhao, X., Yan, X., Yu, A., Van Hentenryck, P. (2020). Prediction and behavioral analysis of travel mode choice: A comparison of logit models and machine learning. [Download Preprint]. Travel Behavior and Society, 20, 22-35. https://doi.org/10.1016/j.tbs.2020.02.003 (Won the 2020 Outstanding Paper Award)
As ridesourcing continues to grow in popularity, being able to accurately predict the demand for it is essential for effective land-use and transportation planning and policymaking. In this research project, we show that using random forest, a machine learning model that can automatically capture nonlinear relationships and interactive effections, can significantly improve predictive accuracy compared to the traditional statistical model.
Key words: random forest, machine learning, travel demand prediction
Yan, X., Liu, X., Zhao, X. (2020). Using machine learning for direct demand modeling of ridesourcing services in Chicago. [Download Preprint]. Journal of Transport Geography, 83, 102661.