Exploring Shape Grammar Optimization as a Tool for Automated Design

Abstract

Shape grammars are quite effective at representing the structure of objects, thus raising the question of whether they could be utilized for design automation or related techniques like procedural generation. However, refining these grammars requires tediously adjusting its many hard-coded parameters. This research serves to answer whether a sub-optimal shape grammar could instead be adjusted using grammar induction and optimization techniques. A general optimization framework for shape grammars was defined to address this question. From this framework, a specific optimization process was also created and its effectiveness was tested in a pilot experiment. To carry out this experiment, a program was written to take a textual design grammar as input and, after several rounds of training by the user, adjust the grammar’s parameters such that it outputs higher-quality designs. After collecting data on the mean grammar design quality per round, it was found that the quality of the designs was, in fact, significantly higher in later rounds than in earlier rounds. This provides an encouraging first step into the potential for applying this optimization framework to design grammars in general.