Inverse Generative Design: A Guideline

Through growing built forms via pre-programmed rules and parameters, the idea of generative design is formed in this modern architecture, in which designers create form alternatives by adjusting input parameters. On the other hand, the emergence of computational platforms asks architects to put more attention to the liveability and sustainability issues of the built environment. While ex post evaluating design alternatives is feasible, it costs much time, crowds out valuable memory, and makes it hard for designers to arrive at optimality. By treating functions as observed variables, this study follows the “inverse problem” philosophy and proposes a novel inverse generative design workflow by finding the latent form variants that lead to these functions. It contains three steps: the first “Generation” randomly generates variants satisfying certain demands; the second “Selection” identifies a temporarily optimal layout given criteria; and the third “Augmentation” finds the major characteristics of this layout and uses it as new knowledge to augment the generator in the first step. This process will continue iteratively until one satisfying layout is found. With an urban renewal case study of To Kwa Wan, Hong Kong, this workflow demonstrates strong practicality by cooling the ambient outdoor environment by 0.10 °C and increasing 1.01% more sky views to the pedestrian.