Rule-Based Reasoning in Dynamic Knowledge Transformer-based Intrusion Detection System for Autonomous Vehicles

In recent years, Autonomous vehicles (AVs) require robust reasoning mechanisms to ensure safety and efficiency dynamic traffic environments. However, existing Hybrid Intrusion Detection System using Deep Reinforcement Learning (HIDS-DRL) model limited by static rule hierarchies, complications in resolving rule conflicts, and lacking generalization to unseen situations. To overcome these limitations, this research proposes the Rule-based Reasoning in Dynamic Knowledge Transformer for AV (RDK-AV) framework that combines symbolic reasoning and data-driven learning. Initially, a dynamic knowledge base encodes traffic laws, contextual constraints, and safety priorities, updated continuously with environmental perception. After that, a rule-based reasoning layer filters candidate actions and resolves conflicts adaptively, while a knowledge-infused transformer captures temporal traffic patterns and aligns them with rule embeddings to support contextual reasoning. Then, a decision fusion module integrates both outputs to generate safe, explainable and adaptive driving policies with increased robustness for autonomous navigation in real-world scenarios. Experimental results demonstrates that the proposed RDK-AV model achieved fairly good performance in optimizing the whole traffic system with Average Cumulative Delay (ACD) per vehicle of 2.6%, when compared with existing HIDS-DRL model.