Parameter-Free Similarity Control: Replacing Fuzzy Membership Functions with Data-Derived Similarity Weights

We present a parameter-free alternative to fuzzy logic controllers based on the Similarity-Induced Method (SIM), where fuzzy memberships are replaced by data-derived similarity weights over prototypes. The resulting controller preserves the structure of fuzzy inference — a normalized weighted aggregation of local rules — while eliminating membership thresholds and shape parameters. We evaluate on a canonical two-input control task (temperature and humidity to fan speed) across multiple severities of distribution shift with 10 random seeds. SIM consistently outperforms both a hand-designed fuzzy controller and a data-retuned fuzzy upper bound. Retuning improves in-domain performance (R² = 0.844) but collapses under strong shift (R² = −0.224), while SIM achieves R² = 0.876 in-domain and R² = 0.639 under strong shift.