This project focuses on developing and evaluating advanced runtime monitoring and decision-making frameworks for intelligent transportation systems and autonomous driving. The primary goal is to design methods that ensure safe and adaptive behavior of autonomous agents operating in uncertain and dynamic environments. To achieve this, I will combine multi-armed bandits (MAB), reinforcement learning (RL), and large language model (LLM)-based reasoning into an integrated pipeline for policy optimization and monitoring. A key challenge in autonomous driving and traffic management is the ability to continuously adapt to changing contexts while maintaining safety constraints. Classical RL approaches can achieve high performance, but they often lack robustness when deployed in real-world conditions. Runtime monitoring is therefore critical: by observing agent behavior in real time, deviations from safety specifications can be detected early, allowing corrective measures to be applied. MAB methods will be explored to balance exploration and exploitation under limited feedback, while RL will be used to learn long-horizon sequential decision-making policies. In addition, recent advances in LLMs will be leveraged to provide higher-level reasoning and explainability, potentially enabling natural language descriptions of decision-making processes and safety violations. The project requires realistic and varied simulation environments. I will use SUMO to model large-scale traffic flows and study adaptive monitoring under congestion, high-density, or stochastic driving behaviors. In parallel, CARLA will be used to test vehicle-level decision-making and safety monitoring in complex urban driving scenarios. By advancing techniques for runtime monitoring and adaptive decision-making, this project contributes toward safer and more trustworthy AI for autonomous driving and intelligent transportation systems. The results will be relevant not only for transportation research but also for broader AI safety applications where runtime monitoring and adaptive control are required.