STPA Enhanced Safety Study for AI-Based Automatic Emergency Braking System: A Combined Approach with Cause Tree Analysis

The increasing complexity of software and the integration of learning systems within the automotive industry demand novel analytical approaches that extend beyond conventional functional safety methodologies. The development of ISO/PAS 8800:2024 was initiated to address the safety implications of uncertainties and model inadequacies present within AI-based systems. This comprehensive framework is centered on ensuring the safety of AI systems. In this study, a safety analysis is performed for an AI-based Automatic Emergency Braking System (AEB) using the System-Theoretic Process Analysis (STPA) method. The identification of unsafe control actions and the subsequent structuring of the causal scenarios corresponding to each one are paramount. Furthermore, the Cause Tree Analysis (CTA) method recommended in ISO 21448:2022 (SOTIF) standard is integrated with the STPA process to obtain a more robust and causally traceable analysis structure. In light of the performance limitations and perception insufficiencies inherent to AI-based decision mechanisms, a set of system measures have been delineated to mitigate potential risks and limit adverse consequences. This integrated approach facilitates the systematic derivation of causal scenarios from unsafe control actions and enables a structured evaluation of risks associated with learningenabled systems, both during design and operation.