Interior design is a labor intensive, self similar task which is difficult to master and cannot be automated using heuristics. This project develops an automated furnishing method that utilizes CGANs for generating detailed architectural layouts from skeletal plans. For this purpose we have developed an architectural tagging method and used it to construct a unique dataset of tagged architectural designs which are used for training the models.
“Soft” fabrication methods, such as formless molding and CNC knitting, rely on complex physical models which are difficult to simulate. This makes “reverse engineering” – determining the fabrication sequence from the desired shape – almost impossible using conventional methods. This project explores the potential of ML to learn from previous experience to predict the needed fabrication steps to arrive at a given geometry.
Building Information Modelling (BIM) – the state of the art architectural format – can be converted into a graph. We will use this to train GNN models to predict the properties of architectural elements according to the subset of their nodes. This will allow us to use ML for checking and auto-completing BIM models, saving time and preventing errors in the architectural design process.
