Engineer develops innovative approach to designing safer streets for cyclists

Roadway design is a complex task. Designing a new road or redesigning an existing one requires addressing a complicated mix of design manuals and standards, traffic impacts and input from stakeholders and local government. 

But there is a key group of stakeholders whose needs are often inadequately considered in road infrastructure, travelers who are more vulnerable to traffic hazards and patterns — cyclists. 

 Xiang Jacob Yan, Ph.D., is trying to change that.  

An assistant professor in the University of Florida’s Department of Civil & Coastal Engineering, or CCE, Yan’s Just & Green Transportation Lab is developing a tool to help designers incorporate cyclists’ perspectives with the help of large language models, or LLMs.  

Part of a larger project funded by the National Science Foundation, StreetDesignAI is an interactive tool that enables designers to generate hypothetical street designs on top of real-world street view images. The system first generates meaningful feedback on the current design using LLM-based AI agents that represent different cyclist populations (“virtual cyclists”). Then, as the designer modifies the design, these virtual cyclists provide updated feedback, including pros and cons of the design concept. 

Skeptics may question the validity of perspectives provided by AI-based virtual cyclists. To ensure the feedback is relevant and meaningful, researchers used a mix of crowd-sourced data, surveys and bikeability assessment data from real cyclists to train these AI agents.

For street designers, it’s like having a group of cycling advocates in the room while they work.  

“Having this system allows the cyclist population to participate in the process and voice their preferences, their needs and concerns,” explained Yan. “When designing roads, roadway engineers have often not considered cyclists’ perspectives very well.” 

Not just one kind of cyclist, either.  

Yan and his co-authors utilize an LLM-powered model based on four types of cyclists, or personas, to take into account the wide variety of biking styles, attitudes and behaviors. Designers receive feedback on their designs from “Strong and Fearless,” “Enthused and Confident,” “Interested but Concerned” and “No Way, No How” cyclists. The tool also considers the driver’s perspective on the proposed design. 

“Confident riders would cycle along with traffic,” Yan explained. “But cautious riders have no desire to be next to cars. So, with these kinds of conflicting needs, roadway designers need to navigate trade-offs when they design. And they have to ask themselves, ‘How will the drivers react?’ There are many pros and cons.” 

The work, “StreetDesignAI: Broadening Designer Perspectives Through Multi-Persona Evaluation of Cycling Infrastructure,” is published in the Proceedings of the 2026 Association for Computing Machinery (ACM) Designing Interactive Systems Conference. 

Yan,serves as principal investigator and mentor for the project, while Ziyi Wang, a graduate student from the University of Maryland, is the lead researcher. Others working on the study include UF graduate student Duanya Lyu and undergraduate student Mateo Nader; Yilong Dai, from the University of Alabama; Sihan Chen, from Carnegie Mellon University; and Wanghao Ye and Zijian Ding, from the University of Maryland. 

The tool features deep analysis for more in-depth results from the virtual cyclists, which provides a sort of dialog with cyclist personas and allows designers to ask follow-up questions such as “Why do you feel unsafe?” or “What changes would improve your assessment?” 

For Yan, this is more than just an academic pursuit — he’s an avid cyclist himself.  

“I'm kind of between a cautious and a confident rider, depending on the context,” he said. “It depends on if I ride my e-bike or my regular city bike. When I’m on my e-bike, I can accelerate much faster, and as the speed gets faster, I feel more confident.” 

Researchers included 26 transportation professionals in their study and gauged their perception of the tool in comparison to general-purpose AI chatbots. They found that the professionals using StreetDesignAI reported significantly higher overall satisfaction with their design work and a marked preference toward using the system as compared to a general chatbot. Anecdotally, respondents said the tool facilitated optimizing designs to include a number of different users’ perspectives. 

As transportation planners and engineers attempt to prepare for an e-bike future, Yan’s research is helping to plan a safer biking infrastructure for all types of cyclists — whether nervous or confident, e-bike or conventional.