Wang, Shuai; Jiang, Emmanuel; Jun, Youngsang
Students of Stuart Weitzman School of Design, University of Pennsylvania
Advisors: Dr. Lin, Zhongjie; Tang, Ziyi; Yi, Shengao
Robotaxi adoption is no longer hypothetical. This study provides visualization and modeling that show how built environments shape robotaxi crash risks in San Francisco, allowing planners and policymakers to implement a "Try-Before-You-Build” approach.
First, you would want to start by trying the web app, which is built using files located in the css, data, js folders.
The data is divided into two categories: crash data and built environment and sociodemographic parameters (independent variables). The final dataset used in the web app is df_final_Sci3.geojson, which includes all dependent and independent, as well as prediction values for each block group.
The crash data was originally obtained as individual incident reports in PDF format from California DMV. These were parsed and consolidated into a single GeoJSON file using parsing.py. It was then uploaded to Firebase using initialcrash.py as the initial dataset.
The dependent variable is crash density, defined as the number of crashes per unit block group area (km²).
By mostly R, all independent variables are compiled and merged in the file CBG 0411_elevationgeometryadded.csv. They are processed and analyzed in 250429_Crash_BE.ipynb.
Variance Inflation Factor (VIF) analysis, Ordinary Least Squares (OLS) regression, and Random Forest (RF) modeling were conducted and visualized in 250429_Crash_BE.ipynb.