An Introspective Approach to Apartment Design
DC I/O 2020 proceeding by PAUL POINET, DIMITRIE STEFANESCU, GEORGIOS TSAKIRIDIS, AURELIE DE BOISSIEU, ELENI PAPADONIKOLAKI.
Contents
Abstract
This paper outlines the development of a residential layout recommendation tool at Allford Hall Monaghan Morris, a large architecture firm based in London. Using software development and machine learning techniques, the firm developed a plugin for Autodesk Revit to assist architects by allowing them to cross-reference any empty space against the firm’s library of completed apartment designs. Nicknamed ‘Homegrown’, this tool recommends apartment layouts for similar spaces. Any apartment chosen by the user is then automatically reconstructed, adapted and post-rationalised to fit the candidate space.
To further enhance the efficacy of this tool, feedback was collected from the firm’s architects which was then used to refine the recommendation algorithm’s feature weightings using random forest regression.
This paper gives an overview of the tool’s development process and focuses particularly on its spatial analysis algorithm, which was developed alongside AHMM’s architectural design teams. It attempts to answer the question: can architecture firms introspect new residential layouts based on existing designs, rather than generating them from first principles?
Keywords
Design Automation, Recommendation, Space Planning, Architecture, Residential Design, Machine Learning, Random Forest Regression.
Topics
AI-driven development, Architecture, Artificial Intelligence in Design, Assisted Design Decision Making, Construction, Design Knowledge Capture, Design Support Systems, Future of work/unemployment.
