Generative adversarial network-driven interactive simulation modelling for environmental design

This paper presents a cognitive-semantic guided generative adversarial network for automatically generating interactive environment layouts that optimise both visual realism and user experience.By computationally operationalising cognitive load theory, our framework integrates a novel interaction-aware discriminator and a semantic consistency loss, enabling the generator to produce layouts that minimise navigational cognitive load.Validated on the Stanford 2D-3D-Semantics dataset, our model significantly outperforms state-of-the-art methods in functional metrics, achieving an 85.2% navigation success rate, a 13.4% higher mean intersection over union than graph-based methods (68.7% versus 55.2%), and a substantially lower cognitive load score of 0.65.Ablation studies and user evaluations involving 45 participants confirm the necessity of each component and demonstrate a strong preference for the generated environments.This work aims to establish between cognitive theory and generative artificial intelligence for human-centric design.