Split Grammar Evolution for Procedural Modeling of Virtual Buildings

Procedural modelling has been successfully applied to the automatic building generation problem. Among several techniques proposed to tackle the problem of procedural building generation, the use of Split Grammars has increased, even being deployed in commercial CAAD (Computer-Aided Architectural Design) software. This work proposes a technique to optimize Split Grammars using Genetic Algorithm. The main goal is to automatically create grammars that only generate models with certain desirable characteristics, either from a series of manually written grammars or randomly created ones. The proposed technique searches the space of the input grammars´ rules to develop new and better grammars, i.e., grammars that generate models with certain pre-defined features. The proposed technique was successfully applied, as will be shown, to the maximization of symmetry of building facades, leading to the creation of realistic models.