Specification-Driven Evolution of Floor Plan Design

Abstract

Generating floor plan designs is a challenging task that requires from an architect both engineering knowledge and creativity. Various computer-aided design tools are used to improve the efficiency of the design process, the most promising of which are intelligent computational models. In this paper a human-computer interaction based framework for multi-storey houses floor plan design is proposed, where the generation of possible solutions is driven by the evolutionary search directed by the user-defined criteria. The constraints and requirements specified by the user provide the basis for the definition of the requirement-weighted fitness function and can be modified during the evolution process. In the first stage of evolution the layouts for one floor are generated. Floor plans for other floors are generated in the next stage, which allows for introducing additional constraints regarding the position of structural elements (such as load-bearing walls or stairs) that cannot be mutated, and thus adjust these plans to the ones generated earlier. The genotypes of individuals are represented by the vectors of numerical values of points representing endpoints of room walls. This structure allows for representing any rectilinear rooms. A case study of the floor plan design for a two-storey house is presented.