A Robust, Transparent, and Steerable AI Agent System for the Design of Floor Plans

2024, Cornelius Carl, Arindam Katoch, Samuel Losi

Project Gallery

Project Video

00:59

A Robust, Transparent, and Steerable AI Agent System for the Design of Floor Plans

This thesis looks to extend the potential of digital architectural tools, harnessing potentials of technologies built upon the recent developments of artificial intelligence (AI), specifically that of large language models (LLM). Application of the LLM has the potential to accommodate fundamental changes to a geometric script through application of an AI agent system while giving the user more control over its generated content by making the AI’s decision-making process more transparent. It reveals an opportunity to reduce time spent on repetitive tasks and allow users to focus more on conceptual aspects of their work. The goal of this thesis is to create a novel architectural tool that enables a designer to collaborate with generative AI by guiding its outputs rather than blindly relying on them. By tailoring an interface frontend to an LLM-based backend, this research develops a medium of interaction between an architect-user and an AI agent system for software-based, early stage design process collaboration. The software grants each actor the ability to play to their unique strengths: architect as decision-maker and AI as content generator. This ensures a robust quality of output, reveals transparent AI reasoning, and gives the user flexibility aimed to steer the AI without strictly defining the AI’s content generation. The design and creation of this software broadens the potentials of generative AI use in architecture, presenting an application that balances two critical, yet opposed aspects of the architectural design process: precision and creativity.

 

ITECH M.Sc. Thesis Project 2024: A Robust, Transparent, and Steerable AI Agent System for the Design of Floor Plans
Cornelius Carl, Arindam Katoch, Samuel Losi

Thesis Advisers: Katja Rinderspacher, Tobias Schwinn, Ali Nakhaee

Thesis Supervisor: Prof. Achim Menges
Second Supervisor: Prof. Jan Knippers, Prof. Thomas Wortmann

To the top of the page