Application of Generative design in Urban Planning

Abstract

Generative design is reshaping urban planning, becoming a major theme in modernisation; however, many aspects of this development are disjointed, with many methods and application areas scattered across the field. Performing this systematic review of literature, we integrated the adoption of generative techniques showing the transition from initial parametric rule-based systems to recent deep learning models, and found the main topics in the publications. Following PRISMA rules, our focused search and screening procedure resulted in 175 peer-reviewed papers that were classified according to eight main areas of investigation. The data shows a distinct move from using simple parametric rules to designing urban forms through probabilistic generative adversarial networks (GANs) and diffusion models; the considerable amount of performance-based design literature focusing on optimising energy consumption, environmental quality and mobility; and participatory planning gradually adopting human-in-the-loop generative systems, but scalability remains problematic. The use of large language models (LLMs) for policy simulation is an initial area that is growing very fast. Unfortunately, the problem of measuring effectiveness and considering ethics, such as bias, transparency, and accountability, is very weak across this research field. This paper posits that even when generative design can automate and improve urban planning processes to a great extent, substantial problems still exist in the lack of standard evaluation frameworks, ethical guidelines and the genuine interweaving of human agency. Future research should therefore prioritise robust validation methods and participatory governance models to ensure that generative urbanism delivers equitable and sustainable urban outcomes.

Keywords: Generative Design, Urban Planning, Deep Learning, GANs, Diffusion Models, Participatory Planning, Evaluation Frameworks

DOI: 10.54941/ahfe1007969