University of Washington Engineers Develop AI Driven BikeButler App for Custom Seattle Cycling Routes

University of Washington researchers launch BikeButler, an AI tool that lets Seattle cyclists customize routes based on greenery, speed limits, and road quality.

By: AXL Media

Published: Apr 29, 2026, 10:37 AM EDT

Source: Information for this report was sourced from EurekAlert!

Addressing the Limitations of Conventional Mapping

For many urban commuters, traditional mapping applications often fail to account for the nuanced realities of cycling. Common platforms frequently suggest routes based solely on the presence of bike lanes, regardless of high traffic speeds, steep inclines, or poor road conditions. Jared Hwang, a doctoral student at the University of Washington, initiated the development of BikeButler after finding that existing tools and crowdsourced advice on platforms like Reddit were insufficient for finding safe and enjoyable routes in Seattle. The resulting application aims to consolidate disparate data sets into a single, user-friendly interface.

Customizing Urban Travel Through Interactive Sliders

BikeButler distinguishes itself from standard GPS services by offering users a high degree of personalization through eight distinct attribute sliders. Cyclists can adjust their preferences for factors such as pavement quality, the amount of surrounding greenery, and local speed limits. As the user moves these sliders, the app generates specific route options that align with their priorities for that particular trip. This allows a rider to choose a fast, direct path for a work morning or a more scenic, protected route for a weekend excursion with family.

Leveraging Artificial Intelligence for Subjective Data

To populate the app with detailed information, the research team combined objective data from OpenStreetMap and government records with subjective analysis powered by a Visual Language Model. This AI tool analyzed thousands of Google Street View images to rate attributes that are not typically found in traditional databases, such as the aesthetic quality of a street or the condition of the asphalt. During testing, the AI’s ratings for greenery agreed with human researchers about 60% of the time, a level of consistency that matches the agreement between human evaluators themselves.