Operations Research and Combinatorial Optimization for 3D Interior Layout Generation

This article explores the application of operations research and combinatorial optimization methods to the automatic generation of 3D home interior layouts, and surveys recent AI‑driven approaches.

Interior design is framed as a large‑scale decision problem: the 3D layout space grows exponentially with the number of furniture items, leading to high computational complexity. The problem is classified among P, NP, NP‑complete, and NP‑hard categories, highlighting its inherent difficulty.

The layout task is modeled as a combinatorial optimization problem. Decision variables represent the 3‑D positions (and orientations) of furniture pieces. An objective function minimizes a weighted sum of constraint violations, including collision avoidance, wall‑adjacency, spatial bounds, and other design rules.

An iterative optimization procedure, analogous to neural‑network training, is described: initialize positions, update constraint weights, compute energy (loss), apply early‑stopping criteria (global loss thresholds), and extract the best solution after multiple rounds.

Two recent AI models are presented. LEGO‑Net treats layout refinement as a denoising diffusion process, using a transformer backbone to predict adjustments from a chaotic initial arrangement. CC3D generates 3‑D scenes from a single‑view image and a 2‑D layout map via StyleGAN2 and neural radiance fields, ensuring consistency between generated and reference layouts.

Experimental results demonstrate that both methods can resolve collisions and produce plausible furniture arrangements, yet fully automatic, high‑quality interior design remains an open research challenge.

The article concludes with an outlook on future AI‑assisted interior design and cites three key references.