Balancing Safety and Security in Autonomous Driving Systems: A Machine Learning Approach with Safety-First Prioritization

In the era of autonomous vehicles (AVs), ensuring the safety and security of the system is an ongoing challenge, particularly when faced with increasingly cyber-attacks such as GPS spoofing and man-in-the-middle. This paper presents a novel reinforcement learning (RL)-based framework that prioritizes safety over security concerns in AVs, ensuring that life-critical tasks such as collision avoidance and speed control are always maintained, even in the presence of security threats. Our system leverages the MAPE-K loop to dynamically adapt to changing conditions, using RL in the Analyzing phase to make real-time decisions that maximize cumulative safety rewards while addressing security risks that directly affect operational integrity. We demonstrate how our RL-based self-adaptive system mitigates security threats while maintaining safe driving behaviors through various simulations and experiments. The proposed system shows significant improvements in collision avoidance and threat detection, ultimately ensuring that the AV’s safety is preserved in even the most challenging scenarios.