Comparison of fuzzy rule-based policy creation algorithms

Decision models based on artificial intelligence need to be validated/certified in order to be integrated into critical systems. Unfortunately, the most efficient models currently are based on neural networks and these are in difficulty with the criteria of explainability and interpretability, necessary for validation/certification. The combination of learning techniques with fuzzy logic is a very interesting avenue for overcoming this problem and thus obtaining a better performance/interpretability compromise. In this paper, we propose a comparison of three methods from the literature on toy cases of control problems by looking at both the performance of the learned control models but also their ability to satisfy interpretability criteria.