A Markov Decision Process Workflow for Automating Interior Design: Fid-Bau Portal

A Markov Decision Process Workflow for Automating Interior Design

Karan, Ebrahim / Asgari, Sadegh / Rashidi, Abbas

We present a novel workflow based on artificial intelligence (AI) techniques to automate interior design processes. We discuss the essential steps for creating an intelligent agent that can automatically perceive the design environment (perception) and produce design ideas (action). In the first step, we use photographs or video images to model three-dimensional coordinates and exact positions of surface points on objects inside the interior space. We then convert the collected spatial data to a set or cloud of points. To fully model the interior space, we create either a triangulated surface or a mesh from the points and then transform it into a detailed building information model (BIM). Last, we apply texture data to either the 3D surface/mesh or the building information model. In the second step, we develop a sequential decision-making model based on Markov decision process for the intelligent agent to make design decisions in the BIM environment. We apply the proposed workflow to a case study with 512 possible design options, conduct experiments with 20 participants where design decisions are made based on AI insights, and perform statistical analysis on the experiment results. Our findings show the proposed workflow is capable of improving participants’ satisfaction by only searching through on average 5.1% of all possible design options. Also, across all performance measures, design decisions proposed by the AI system outperform designs made randomly.