HuVA Lab · KSAE Korean Automotive Journal
California AV disengagement reports — HMI-focused longitudinal study
Context
Conducted a 10-year longitudinal study of California DMV autonomous-vehicle disengagement reports, integrating heterogeneous PDF and Excel disclosures from 54 manufacturers. Reframed disengagement analysis as a Human–Machine Interaction problem by examining not only failure causes, but also disengagement initiators and long-term driver–AV interaction patterns. Published as first author in the KSAE Korean Automotive Journal, with an IJAT follow-up paper in preparation.
KSAE Spring Conference — Live Presentation
Delivered the methodology and findings live to automotive-safety researchers at the Korean Society of Automotive Engineers (KSAE) Spring Conference — walked through the Driver / System / Environment taxonomy, the 2018 Apple/Uber-ATG outlier handling, and the HMI implications of the 69–88% preemptive-intervention finding.
KSAE Conference Q&A — Defending the Framework
Fielded technical questions from automotive engineers on the classification framework, the longitudinal trend interpretation, and the Tesla Level-2 regulatory exemption finding — the discrepancy between public perception of “Tesla Autopilot = autonomous driving” and the CA DMV classification that exempts it from disengagement reporting.
Full Paper Accepted — KSAE Korean Automotive Journal
Authored the first-author paper “A Decade of California Autonomous Vehicle Disengagement Reports: HMI Perspective and Longitudinal Trends.”Standardized 222,537 disengagements across 54 manufacturers (2015–2024) into a unified Driver / System / Environment schema; extracted tabular & narrative data from heterogeneous DMV PDFs using Tabula, Camelot, Excel VBA, and Python. Key findings: 2018-outlier-excluded analysis shows System factors at 72.8%; 69–88% of all disengagements were preemptive safety-driver interventions, motivating trust-building HMI design rather than purely technical hardening.
IJAT Follow-up — Multilingual Transformer Extension (in preparation)
Drafting a sole-first-author IJAT follow-up on cross-manufacturer fault attribution in California AV disengagement reports. Fine-tuned XLM-RoBERTa and pseudo-labeled 48,360 high-confidence narratives across 19 OEMs to quantify manufacturer-level System / Environment / Driver attribution patterns. Found significant cross-OEM heterogeneity (deduplicated Cramér's V = 0.440, p<10⁻⁸³), then mapped OEM profiles into four HMI design hypotheses: trust calibration, handover smoothness, routine-operations communication, and ODD-boundary signaling.