Despite advancements in automated driving technology, traffic accidents remain a significant safety concern in urban areas because of complex traffic environments and current technical limitations of automated vehicles (AV). Structured, quantitative scenario-based evaluations are essential for assessing AV safety. This study aims to generate functional scenarios by accurately inferring vehicle and pedestrian behavior information from unstructured traffic accident descriptions. The dataset comprises 137,432 traffic accident cases collected by the Korean National Police Agency. A total of 1,774 training samples were extracted using term frequency–inverse document frequency keyword analysis. We then evaluated the prediction performance of three large language models (LLM): Claude 3, GPT-4o, and GPT-4o with fine-tuning, using 500 randomly selected traffic accident records that lacked keyword-extractable behavior information. GPT-4o with fine-tuning achieved the highest accuracy of 81.1%. Using this model, 26 functional scenarios were generated. The risk levels for each scenario were evaluated by integrating the equivalent property damage only index with risk predictions from logistic regression and XGBoost models. A weighted average of the normalized results was used to rank the scenarios. Scenarios such as “turning left / crossing,” “straight / crossing,” and “turning right / crossing” were consistently identified as high-risk. This study demonstrates the effectiveness of LLM-based contextual inference in converting real-world traffic accident data into functional scenarios, and contributes to the advancement of AV safety evaluation frameworks.
1 January 2025 in Journal of Advanced Transportation
Understanding the conditions that affect takeover time (TOT) in conditional autonomous driving systems remains a challenging issue. The takeover process requires a seamless transition of control from the autonomous system to the driver when the system encounters situations it cannot manage. This study examines the effects of traffic conditions, road geometry, and weather on TOT using a linear mixed model to quantify their influence. Preliminary findings indicate that factors such as rain, gender, and age significantly extend control transition duration. These insights highlight the need for personalized designs in automated driving systems (ADSs) and takeover request protocols to accommodate diverse driver characteristics and environmental conditions. While the research utilizes a driving simulator, suggesting the need for field validation, it offers a foundational understanding that can enhance the safety and efficiency of conditional automation systems. This study contributes to safer ADS design and supports the commercial viability of conditional autonomous vehicles.