Developing a Comprehensive Dataset and Categorization of Automated Vehicle Crashes
Research Team: Julia Griswold (lead) and Iman Mahdinia
UC Campus(es): UC Berkeley
Additional Research Partners: California Department of Motor Vehicles (DMV)
Problem Statement: Despite detailed crash reports mandated by the National Highway Traffic Safety Administration and California Department of Motor Vehicles, a cohesive, preprocessed dataset for analyzing automated vehicle crashes remains unavailable.
Project Description: To better understand automated vehicle safety dynamics and provide insights to safety improvements, this research project will develop a comprehensive automated vehicle crash dataset using Natural Language Processing and a manual review of the narratives. By employing unsupervised machine learning, the project will also categorize crash scenarios into rare (i.e., strong edge cases), complex (i.e., weak edge cases), and usual types; and analyze human behaviors involved in automated vehicle crashes in mixed traffic.
Status: In Progress
Budget: $100,000