Convolutional Neural Networks (CNNs) are increasingly becoming integral, particularly in the context of Cyber-Physical Systems (CPS), such as autonomous vehicles. The proliferation of CNNs in safety-critical applications introduces a significant challenge due to their complex internal workings, making the verification of such systems inherently challenging. It is of utmost importance that any CNN used in CPS is verified to be safe for that system in all possible environments. Through this work, we propose a formal runtime verification (RV) approach that employs user-specified policies constructed through Valued Discrete Timed Automata (VDTA). This innovative approach serves as a powerful tool for ensuring the adherence of the system to predefined safety-critical policies, providing a formalized method to handle the complexities of CNNs in safety-critical contexts by treating them as black boxes.

To demonstrate the effectiveness of this runtime verification approach, the study focuses on the autonomous vehicle (AV) scenario. We develop CNN models that are trained using RL-based Q-Learning techniques and integrate them into some of its key components. By integrating RV monitors with this system, we demonstrate their potential to increase the safety and reliability of the overall system. We also observe this method to produce minimal overheads. Submitted