| Title |
AI-based Spatial Arrangement Simulator with Reinforccement Learning |
| Authors |
이상현(Lee, Sang-Hyun) ; 지성운(Chi, Cheng-Yun) |
| DOI |
https://doi.org/10.5659/JAIK.2021.37.11.43 |
| Keywords |
Artificial Intelligence; Artificial Neural Network; Architectural Design; Reinforcement Learning; Deep Q-Network; Spatial Layout |
| Abstract |
The purpose of this study is to develop reinforcement learning-based spatial layout simulators. A spatial layout simulator means placing the
unit spaces that make up the entire architectural space in the appropriate location and in the appropriate neighborhood relationship when
given. In this study, we conducted 1) architectural design process analysis 2) simulator development 3) validation of simulator. As a result of
architectural design process analysis, it was confirmed that the layout of the architectural interior space is essential. Simulator development
includes 1) establishment of reinforcement learning environment 2) implementation of agent 3) implementation of reward system. To validate
the simulator, three planar types were presented; a reward scheme was devised to guide each type; and the simulation was conducted
according to the reward scheme. The simulation resulted in the desired planar type being produced by controlling the reward scheme (without
having to learn the knowledge and experience of human architects). This confirms the availability of reinforcement learning-based spatial
layout simulators. |