Yang-Jun (Andrew) Joo, Ph.D.
Autonomous Driving Safety Evaluation · Conformal Prediction · Traffic Safety Analytics
I develop data-driven methods to evaluate driving safety, calibrate AI/ML uncertainty, and analyze large-scale mobility datasets. My research combines traffic safety engineering, deep learning (including vision-language models), conformal prediction, and big data pipelines to support safer connected and automated transportation systems.
Research Focus Areas
Key domains at the intersection of traffic engineering, machine learning, and vehicle safety.
Autonomous Driving Safety Evaluation
Developing human-aligned risk perception metrics and evaluating driverless AV passenger risk judgment using naturalistic videos and surveys.
Uncertainty Calibration & VLMs
Applying conformal prediction to calibrate vision-language models (VLM) for driving scene QA, ensuring reliable confidence bounds in safety-critical settings.
Large-Scale Mobility Data Pipelines
Designing spatial-temporal data workflows and dashboards to process high-resolution traffic signal logs (ATSPM) and vehicle telemetry.
Traffic Safety & Risk Analytics
Utilizing risk field theory and surrogate safety measures (SSMs) to model rear-end conflicts and identify micro-level safety hotspots.
Featured Projects
Highlighting applied research and technical developments in autonomous safety and traffic intelligence.
Assessing Perceived Risk Of Autonomous Vehicle Passengers Using Dashcam Crash Video
Autonomous driving systems need to align with human risk perception in near-miss and safety-critical scenarios to improve passenger trust.
Completed PI project funded by National Research Foundation of Korea (73.5M KRW), yielding publication manuscripts (under 1st revision at IEEE T-ITS).
Enables AV developers to align vehicle behaviors with passenger risk expectations and design trust-building passenger protocols.
Reliability-Based Risk Assessment Method for Autonomous Driving Using Surrogate Safety Measures
Proactive risk assessment frameworks are needed to evaluate autonomous vehicle safety in real time without relying solely on historical crash records.
Completed PI project funded by National Research Foundation of Korea (96.1M KRW), culminating in doctoral thesis and publications in AMAR and AAP.
Provides safety diagnostics algorithms to run real-time risk evaluation pipelines and calibrate proactive AV collision-avoidance controls.
Network Level Proactive Traffic Operations Indicator (NPTOI) Using Artificial Intelligence (AI) and Connected Vehicle Data Integration
Urban transportation networks lack proactive indicators that forecast traffic states and systems performance in real time using sparse connected vehicle telemetry.
Active research project yielding manuscripts under review at IEEE T-ITS and TRB.
Provides traffic management centers and fleet routing systems with highly accurate network-wide traffic and speed predictions.
Prediction of Safety Opportunities Using Automated Traffic Signal Performance Measures
State transportation agencies need automated ways to process high-resolution signal event records to evaluate corridors and identify safety opportunities.
Completed FDOT-sponsored research project (2024–2025) and agency-facing analytical dashboard.
Demonstrates capability in building scalable data pipelines that convert raw hardware event logs into actionable agency decisions.
Technical Strengths
Core competencies translated from rigorous academic research into industry-ready workflows.
Data & Programming 6 Skills
Machine Learning & AI 6 Skills
Transportation Safety & Mobility 6 Skills
Research Translation & Communication 6 Skills
Selected Publications
Peer-reviewed research in top-tier intelligent transportation systems and computer vision venues.
Context-Aware Conformal Prediction for VLM-Based Driving Scene QA
YJ Joo, Y Kim, SY Park, M Abdel-Aty
Applies conformal prediction to calibrate vision-language models (VLM) for autonomous driving scene question answering, guaranteeing reliable confidence bounds in safety-critical settings.
Provides safety-critical AI systems with calibrated uncertainty estimation, reducing hallucination risks in autonomous vehicle decision-making.
A generalized driving risk assessment on high-speed highways using field theory
YJ Joo, EJ Kim, DK Kim, PY Park
Applies artificial potential field theory to model dynamic driving risk on high-speed expressways, capturing interactions between the vehicle, road geometry, and traffic flow.
Serves as a foundation for real-time collision-avoidance systems and adaptive cruise control safety wrappers on highways.
A data-driven Bayesian network for probabilistic crash risk assessment of individual driver with traffic violation and crash records
YJ Joo, SY Kho, DK Kim, HC Park
Develops a probabilistic Bayesian network to assess individual driver crash risk by combining historic traffic violation records and crash demographics.
Enables auto insurance underwriters and commercial fleet managers to build personalized risk profiles and identify high-risk drivers.
Evaluating the Effect of Signal Adjustment on Rear-End Conflicts Using High-Resolution Event-Based Data
YJ Joo, M Abdel-Aty, BMTH Anik, Z Islam
Leverages high-resolution event-data (ATSPM) to model how traffic signal timing adjustments impact rear-end conflict rates at signalized intersections.
Allows city traffic managers to tune signal timings to simultaneously optimize flow and safety.
Professional Timeline
Postdoctoral and doctoral positions at leading transportation research centers.
Postdoctoral Scholar (Quantitative Safety Analytics)
University of Central Florida
- Led cross-functional safety analytics workstreams for the Florida Department of Transportation, defining and standardizing safety-performance metrics from connected-vehicle and naturalistic driving data.
- Secured KRW 73.5 million in competitive funding as principal investigator and used causal forests to quantify heterogeneous effects of vehicle automation on risk perception across 926 participants and 1,325 near-miss clips.
- Built and automated Python and SQL ETL and analytics pipelines for multi-terabyte datasets from 1,000+ intersections, standardizing risk-metric generation and recurring reporting.
- Developed uncertainty-aware evaluation workflows for autonomous-driving vision-language models using conformal prediction, identifying context-specific calibration failures in safety-critical driving QA.
Postdoctoral Researcher (AV Safety Systems)
Seoul National University
- Led research on real-time infrastructure-to-vehicle (V2I) safety information systems for the Korean National Police Agency.
- Defined safety-critical scenarios and operational safety criteria for infrastructure-assisted autonomous-vehicle safety evaluation.
- Integrated physics-based conflict metrics into analysis workflows to systematically evaluate infrastructure-assisted AV behaviors.
Graduate Research Assistant
Seoul National University
- Developed generalized quantitative risk-assessment methods for autonomous-vehicle lane changes and collision avoidance using field theory, Bayesian networks, reliability analysis, and surrogate safety measures.
- Secured and led a KRW 96.1 million Global Ph.D. Fellowship, developing and validating probabilistic AV risk-assessment methods for lane changes, collision avoidance, and driver-level crash risk.
- Published peer-reviewed research on autonomous-vehicle risk assessment, probabilistic crash risk, and traffic-conflict analysis.
Education
Ph.D., Civil & Environmental Engineering (Transportation), 2023
Seoul National University
B.S., Civil & Environmental Engineering, 2018
Seoul National University
Interested in autonomous driving safety, mobility AI, or large-scale transportation data analytics?
Let's connect. I am open to applied researcher and data science roles within safety engineering, risk analytics, and smart mobility domains.