UC Berkeley SafeTREC is excited to be a part of the Center for Pedestrian and Bicyclist Safety (CPBS), a Tier-1 University Transportation Center (UTC) supported by the United States Department of Transportation (USDOT) and led by the University of New Mexico (UNM).
US DOT welcomes members from new UTC’s at the University of New Mexico, City College of New York, and University of Texas Rio Grande Valley (Photo: US DOT)
CPBS’s goal is to eliminate pedestrian and bicyclist fatalities and serious injuries via research, education, technology transfer, and workforce development. The UC Berkeley team, led by SafeTREC Director Julia Griswold, is one of five partnering institutions, along with the University of New Mexico, San Diego State University, University of Tennessee, Knoxville, and University of Wisconsin-Milwaukee.
CPBS focus areas
- Research
- Education
- Technology Transfer
- Workforce Development
CPBS research in progress at SafeTREC
Identifying Harsh Driving Behaviors and Contributing Factors Using Telematics Data: A Case Study in Oakland and Fresno, California
Principal Investigator: Julia Griswold, University of California, Berkeley
Despite extensive safety countermeasures, vulnerable road users continue to face significant risks on urban roadways, resulting in a substantial loss of life. Safety frameworks like Vision Zero and the Safe System Approach call for proactive solutions that address these dangers before severe crashes occur. This proactive approach can be powered by surrogate safety measures, which use data on near-misses and risky behaviors to identify hazards. Harsh driving events—such as harsh braking or acceleration—serve as excellent indicators of elevated crash risk. These behaviors are influenced by a combination of factors, including roadway design, traffic flow, and the complex, unpredictable interactions between vehicles and other road users in dense urban environments. This study leverages high-resolution telematics data from the Cities of Oakland and Fresno to investigate the differential impacts of harsh driving behaviors on road safety.
We will construct and compare crash hotspots (e.g., high injury network) and harsh driving behavior hotspots to examine which types of harsh driving behaviors most strongly align with crashes involving vulnerable road users and latent crash risks. Additionally, by using statistical methods and explainable artificial intelligence techniques, we will analyze roadway characteristics (e.g., intersections, lane curvature, or slope), as well as traffic flow and surrounding conditions, to determine whether specific features are associated with increased prevalence of harsh driving events that, in turn, elevate crash risk.
By integrating the spatiotemporal patterns of crashes and telematics-based behavioral measures, along with infrastructure characteristics, this study aims to better understand how risky driving patterns contribute to vulnerable road user safety outcomes. The findings will provide actionable insights for prioritizing enforcement such as speed camera deployment, designing infrastructure countermeasures, and developing data-driven, proactive strategies to support safe transportation. Learn more about the 25UCB01 research project.
Are Automated Vehicles Safer Drivers than Humans? Comparing Performance in San Francisco
Principal Investigator: Julia Griswold, University of California, Berkeley
This research project seeks to determine if automated vehicles (AVs) are safer drivers than humans by comparing their pedestrian interaction behaviors and yielding performance in real-world conditions in San Francisco. The study will be framed by the city's "Focus on Five" strategy, which targets the five moving violations most commonly associated with traffic fatalities. Researchers will conduct evaluations of two focus violations, with the first being a comparison of the compliance of AVs and human drivers in yielding to pedestrians in a crosswalk. To gather data, the team will install high-resolution video cameras at two or more crosswalks with no traffic control for a period of one to three weeks to passively record vehicle-pedestrian interactions. Machine learning-based computer vision methods will then be used to automatically classify vehicles as either automated or human-driven. Following this classification, researchers will review the footage to code each interaction, noting if the vehicle yielded to the pedestrian. Finally, the performance of the two groups will be compared using two-sample t-tests to determine if any observed differences are statistically significant. A parallel analysis will be conducted for a second violation, to be determined. Learn more about the 25UCB02 research project.
Examining Attribution of Fault in Fatal & Serious Injury Crashes between Drivers and Vulnerable Road Users
Principal Investigator: Corinne Riddell, University of California, Berkeley
Traffic crashes killing or severely injuring pedestrians and bicyclists have increased dramatically in the past 10 years in the US. Research has found media and police narratives to play a strong role in shaping public opinion around strategies to prevent crashes involving vulnerable road users (VRUs), including whether those narratives attribute blame to the VRU victim. In California, following a crash, data are collected by law enforcement officers who make a determination of the “party most at fault.” Concerns exist as to whether fault in crash report data is over-attributed to VRUs, but no systematic research has assessed this.
We will utilize 2016-2023 electronically reported California Crash Reporting System data to identify the presence of a crash witness at each fatal or serious injury (FSI) single-vehicle/VRU crash, under the hypothesis that fault is more accurately attributed when a witness is present. Because VRU victims involved in FSI crashes with a motor vehicle are often unconscious, in transport to a hospital, or deceased at the time of the crash investigation, only the driver’s perspective is incorporated into police reports in the absence of a witness, while the presence of a witness provides additional and, we hypothesize, more objective information. The specific research question is: is the presence of a witness associated with a higher probability that the driver will be named at fault in FSI crashes between a driver and a VRU compared to crashes between a driver and a VRU when no witness is present?
We will use logistic regression to estimate the probability of the driver being attributed fault in crashes where a witness was present compared to crashes with no witness. If the probability is higher when a witness is present, we will discuss the likelihood that this implies that fault is systemically over-attributed to the VRU when there is no witness vs. other explanations. We will control for neighborhood and investigate if the effect of a witness on attribution varies according to reporting agency, victim attributes, or victim mode. The overall goal will be to estimate minimum correction factors and 95% confidence intervals for VRU-involved crashes so statistics on attribution of fault can be adjusted for future research or in policy settings. Learn more about the 25UCB03 research project.
CPBS completed research at SafeTREC
Year 2 (2024 - 2025)
Pedestrian Fatalities & Injuries in Hit-and-Run Crashes in California, Tennessee & the US: Recent Trends and Risk Factors
Principal Investigator: Julia Griswold, University of California, Berkeley
Both pedestrian fatalities and overall hit-and-run (HAR) fatalities in the US are at a 40-year high, but no post-COVID trends in fatal HAR pedestrian crashes have been examined despite increased reports of reckless driving, increasing vehicle weight and height, and increases in distracted driving. Further, few studies have examined trends in non-fatal pedestrian HAR crashes. Using 2009-2022 national crash fatality data and data on crashes at all severity levels in California and Tennessee, we will examine time trends in all HAR crashes, all pedestrian victim crashes, and how these are related. We will also examine the risk of serious injury or death among HAR vs. non-HAR crashes to try to elucidate the relationship between HAR and outcome severity. Using regression techniques, we will then examine risk factors for single vehicle-pedestrian crashes, including comparing risk factors for HAR vs non-HAR crashes and predictors of whether drivers are eventually identified in HAR crashes. Factors to be examined include crash characteristics, victim characteristics, and driver/vehicle characteristics, where available. We also plan to examine the joint characteristics of driver-pedestrian pairs, such as by age, race or sex, to understand whether this pairing affects the likelihood of fleeing. Finally, we will examine the effect of several inflection points on HAR crash rates and outcomes for pedestrians, including the effects of the COVID pandemic and, potentially, the effects of specific state-level policy changes around licensing laws, given past research linking HAR to unlicensed drivers. Learn more about the 24UCB01 research project.
Fire Safety and Safe Streets: Understanding Conflicts between Safe Streets Improvements and Emergency Response
Principal Investigator: Zachary Lamb, University of California, Berkeley
In this project we propose to address the following questions related to conflicts between fire departments and safe streets efforts: When and why do fire safety and street safety goals come into conflict? What institutional arrangements, design processes, and other practices are emerging to reconcile these conflicts to improve overall community safety? How might best practices for avoiding conflict and finding synergies be replicated from city to city?
The project includes four main components: 1) assembling a national Community Advisory Committee of 8-12 members from across a range of relevant areas of expertise; 2) a review of existing scholarly and gray literatures on conflicts between bike and pedestrian safety and emergency response; 3) construction of a national database of conflicts between street safety upgrades and emergency response demands from 2010 to 2024; 4) development of 3 to 5 in-depth case studies of street safety / emergency response conflicts drawing on stakeholder interviews, local media, project documentation, and public meeting records. Deliverables include: the conflict database, a research paper, and a public report. Learn more about the 24UCB02 research project.
Leveraging Retrieval Augmented Generation (RAG) to Analyze Crash Reports Narratives
Principal Investigator: Julia Griswold, University of California, Berkeley
Crash reports serve as a vital source of information for understanding road crashes, devising strategies for prevention, and informing policies. However, the coding on these reports often lacks detailed characteristics crucial for comprehensive analysis of pedestrian and bicyclist crashes. Crash reports typically contain structured data, which may lack the nuanced details often found in the narrative section regarding the circumstances surrounding a crash. Information such as unhoused status of a pedestrian, detailed explanation of the vehicle movement before hitting a pedestrian, witness description of a speeding vehicle’s behavior pre-crash, and description of a hit-and-run crash conditions may be embedded within the narrative descriptions but remain unrecorded in the structured fields of the report form. Extracting this implicit data poses a significant challenge for traditional analysis methods. Retrieval Augmented Generation (RAG), employs an embedding model to scan extensive text, seeking similarities between the query—here, the presence of a vulnerability factor or demographic context—and segments of the text. Once relevant portions are pinpointed, both the query and context undergo analysis by a Large Language Model (LLM). In this instance, the LLM validates the presence of and extracts pertinent information. This study will explore the ability of RAG to identify crash characteristics found only in the crash report narratives using crash reports from California. Learn more about the 24UCB03 research project.
Year 1 (2023 - 2024)
A Context-Sensitive Street Classification Framework for Speed Limit Setting
Principal Investigators: Julia Griswold, University of California, Berkeley; Robert J. Schneider, University of Wisconsin-Milwaukee
Historically, speed limit setting (SLS) procedures in many states have relied on driver-behavior-based approaches such as the 85th percentile speed. Researchers, however, have identified several shortcomings, including that drivers underestimate their speeds, issues with speed creep, and the lack of consideration for vulnerable road users. States can move towards a context-sensitive approach to SLS by developing a street classification framework that includes context. New Zealand provides a leading example with its SLS procedure that uses a street category framework based on the Movement and Place principle. We will develop a US street category framework for SLS using objective, publicly available datasets that capture functional classification (movement) and variables associated with vulnerable road user activity (place), such as land use mix, population density, job density, urban/rural designation, and transit access. We will perform a pilot study of 5 geographically diverse states to categorize the streets on their road networks based on the Movement and Place principle. The research will include a best practice literature review, GIS data preparation and linking, application of classification methods, validation of classification, and reporting. Learn more about the 23UCB01 research project. New release: access the full research report now available!
Creating a Data Resource of California Police Stops for Use in Traffic Safety Applications
Principal Investigator: Julia Griswold, University of California, Berkeley
Traffic stops are one the most common ways in which the American public interacts with police. Although one of the leading reasons given for police traffic stops is a violation of the vehicle code, there is limited and mixed research on the impact of traditional police traffic enforcement on traffic safety outcomes. At present, few large data resources with an appropriate level of detail exist to facilitate investigations of this type. The 2015 Racial and Identity Profiling Act (RIPA) requires all law enforcement agencies in California to collect and submit vehicle (including bicycle) and pedestrian stop data to the State Department of Justice annually, starting no later than 2022. This project will use 2018-2022 confidential RIPA stop data to categorize all police traffic stops using known risk factors for fatal and severe collisions and to create new variables relevant to traffic safety, yielding a standardized statewide data set useful for examining and controlling for police traffic stops as they relate to traffic safety outcomes. Further, we will both establish clear guidance for how to process RIPA data efficiently for future data releases and will also geospatially join the processed RIPA data files with traditional transportation and land use data sources using stop location so that this data resource can be made available to others for future research. Learn more about the 23UCB02 research project.
CPBS education at SafeTREC
As part of education efforts at the CPBS, SafeTREC provides student-led research and fellowship opportunities.
Summer 2025 CPBS Fellowship
Nikou Khoshnevis Asl, Department of Civil and Environmental Engineering
A Vision-Based Analysis of Infrastructure Impacts on Pedestrian and Bicyclist Crash Outcomes in California
This study uses computer vision techniques and Google Street View imagery to systematically assess the built environment at locations of historical pedestrian and bicyclist crashes in California. By extracting and analyzing street-level infrastructure features, the research explores how these factors relate to crash severity for vulnerable road users. The findings aim to guide targeted infrastructure investments to improve safety for pedestrians and cyclists.
Summer 2024 CPBS Fellowship
Masuma Mollika Miti, Department of Civil and Environmental Engineering
Bicycle Crash Incidents in San Francisco before, during, and after COVID-19
This study examines how bicycle crash patterns in San Francisco changed before, during, and after the COVID-19 pandemic. It found a drop in crashes during the pandemic that aligned with reduced bicycle counts, with most reductions occurring in high-density areas and on weekdays. Weekend crash rates and lower-density areas saw little change, and the lower crash levels have continued post-pandemic. The findings emphasize the need to incorporate population density into urban safety planning, especially during large-scale disruptions like the COVID-19 pandemic.
Learn more about the CPBS and current research, education, technology transfer, and workforce development activities at the CPBS website.
