Integrating domain knowledge and user requirements in kitchen layout optimization

Abstract Layout design data sources exhibit a fundamental asymmetry: the challenge lies in processing structured, high-quality domain knowledge alongside task-specific natural language user requirements. To address this, we propose a hybrid evaluation framework for evolutionary algorithms (EAs) that quantifies constraint violations as virtual displacement costs. The framework is evaluated in the present study on four kitchen-layout optimization scenarios with task-specific user requirements and component sets. By integrating hard-coded domain rules with a fuzzy-logic-driven Abstract Syntax Tree (AST), the system enables a flexible yet explainable scoring mechanism. This approach allows for a granular penalty system for individuals, while also guiding local search heuristics through component-level diagnostic logging. The implementation of the AST results in a marginal reduction in convergence cost while reducing the relative dispersion of the algorithms. Within this kitchen-layout benchmark, the evolutionary algorithms achieved lower execution times and aggregate penalty scores than the selected LLM chat applications. The three non-expert human-generated layouts were included only as an exploratory reference because of the small and demographically homogeneous participant sample.