Intelligent Urban Design: Self-Organized Block Form Generation Using Reinforcement Learning – An Empirical Study from Nanjing, China

Abstract

With the rapid rise of AI technologies, intelligent design has enhanced urban planning through algorithm-driven diversity, flexibility, and efficiency. This study proposes a reinforcement learning-based method for the self-organized generation of block forms, enabling adaptive spatial layout optimisation. Using Nanjing, China, as a case study, geometric calculations extract topological relationships to build a prototype database. A reinforcement learning model is then applied, incorporating design constraints like development intensity, building density, height, and green space ratio. Generative design experiments on three Nanjing blocks demonstrate how feedback allows dynamic adjustments, achieving economic, ecological, and social benefits. Unlike traditional rule-based approaches, this model learns through autonomous interaction with the environment and can better adapt to environmental characteristics, offering a novel framework and practical support for intelligent urban design.