Bridging the Semantic Gap in Autonomous Braking Systems: A Neuro-Symbolic XAI Approach

Abstract

Contemporary deep learning systems in autonomous vehicles extend beyond speed and accuracy toward interpretability and trustworthiness. Regulators, safety engineers, and end-users increasingly demand transparency in how these systems operate. Deep learning models are helpful in many fields of life for human beings but their inherent black-box nature creates a significant disconnect between the low-level technical data they process and the high-level concepts that humans use to interpret and trust automated decisions. Traditional tools such as LIME and SHAP help us in understanding the models inner operations but are often too slow and produce visuals that are difficult to interpret. This paper introduces a new framework called Neuro-Symbolic explainable AI (XAI) to address this semantic gap, specifically designed for real-time brake light prediction and distance monitoring. We are combining the advanced perception model with distance monitoring sensors, translating technical detections into a formal knowledge base of traffic rules and safety concepts. This mixed approach allows the system to generate natural language explanations that justify vehicle actions, such as ”Warning: Lead vehicle braking at a critical distance.”