| Title |
Optimization of Reward Systems for Spatial Layouts Using Reinforcement Learning Based on Drawing Recognition Models |
| Authors |
지성운(Chi, Cheng-Yun) ; 이세원(Lee, Se-Won) |
| DOI |
https://doi.org/10.5659/JAIK.2024.40.11.111 |
| Keywords |
Drawing Recognition; Spatial Layout; Spatial Information; Adjacency Relationship; Reinforcement Learning; Reward System |
| Abstract |
This study presents a methodology for optimizing the reward system of reinforcement learning-based apartment spatial layout using a drawing
recognition model. The model automatically identifies and quantifies adjacency relationships in graphic apartment floor plans through multiple
stages, including pretreatment, binarization, vectorization, main wall recognition, room recognition, and post-treatment. The extracted adjacency
information is quantitatively evaluated using the spatial analysis indicators, which assesses the necessity for adjacency and integrates it into
the reinforcement learning reward system. The optimized reward system enables agents to efficiently learn and generate optimal spatial
layouts. Simulation results confirm that applying type-specific apartment reward systems produces consistent and efficient spatial layouts. This
study contributes to the advancement of AI-based architectural design technology, providing a foundation for practical and human-centered
spatial layout solutions. |