Optimizing BIM drawing element placement through reinforcement learning

Building information modeling (BIM) enhances communication in the architecture, engineering, and construction industry and automates drawing generation. However, optimizing the placement of drawing elements remains a challenge. This paper proposes a framework using proximal policy optimization to improve BIM drawing element placement, focusing on floor plan-type drawings in the construction documentation phase. Deep reinforcement learning ensures stable performance in high-dimensional, sparse-data environments. Experiments on a dataset of 150 drawings showed that the interference ratio among drawing elements converged to zero within 0.05 s to 5 min, improving readability. Compared with conventional BIM processes, the proposed framework reduced “element position adjustment” time and commands by 93.9 % and 94.3 %, respectively, leading to an overall reduction of 25 % in work time and 17.9 % in commands. These results validate the framework's effectiveness in improving productivity and reducing manual effort. It enhances readability, minimizes human errors, and allows designers to focus on essential tasks.