ScholarWorks@숙명여자대학교

초록

To create a reliable autonomous driving environment, advanced artificial intelligence technologies, such as object recognition and path prediction, are being applied. However, the road environment is unpredictable and dynamic, with real-time challenges. If the focus is only on recognizing objects and predicting their movements and paths, there are limitations in complex situational reasoning, real-time attention, alertness to surroundings, and proactive driving. This paper proposes the Roadmarking Thinking Causality Architecture (RTC Architecture), which enables the derivation of various potential driving scenarios and appropriate responses through the interpretation of the intrinsic meaning of road markings. The RTC Architecture employs the CLIP model to transform road marking images into interpretable text. The text-based road markings are processed by a GPT-based model, which generates complex scenarios and driving response strategies for each, allowing for the derivation of the Road Marking Thinking Causality to explain the causal relationship between road markings and potential outcomes. Rather than relying on the movement of dynamic objects, RTC Architecture infers driving situations through the interpretation of the intrinsic meaning of road markings, allowing for the extraction of complex information from the driving environment. By understanding the causality between potential scenarios and corresponding response strategies from road markings and interpretable explanation, RTC Architecture supports reliable decision-making and recognizing potential risks proactively vigilant driving.

키워드