RAFAEL: A Rational Fuzzy Approach for Explainable, Adaptive and Safe Decision-making under Uncertainty

Ensuring that safety requirements remain interpretable and traceable from design-time analysis to run-time decision-making remains a key challenge in safety-critical autonomous systems. This paper proposes RAFAEL, a decision-making approach that preserves stakeholder driven linguistic reasoning throughout the system lifecycle. Grounded in rational decision-making, RAFAEL encodes concepts such as safe and risky into a hierarchy of fuzzy inference systems, enabling graded reasoning under uncertainty and adaptive trade-offs between safety and performance. Evaluated in a car-following scenario, RAFAEL maintains stronger safety margins in critical conditions, achieves comparable efficiency, and produces smoother control actions than a threshold-based baseline. It also demonstrates robustness to noise and design-time uncertainty, highlighting the value of linguistic representations for interpretable and reliable decision-making.