TR044: An Approach to Automated Architectural Floor Layout Generation with Case-Based Reasoning

Research into automated floor layout generation has been a problem in which many Architectural / Engineering / Construction (AIEIC) researchers have taken an interest since the 196Os. These researchers have applied many approaches to solve this problem, however, few results which are considered acceptable by architectural designers have been achieved. Since the approaches required quite a few restrictions for analysis, or were focused on a narrow problem domains, the assumption has been that automated layout generation is still not practical for designer's purposes.

The previous approaches were based on mathematical methods dependent on the processing ability of specific computers. In contrast, a recent approach to the layout generation problem has evolved incorporating know ledge-based paradigms such as rules and frames. Teicholz and Chinowsky (Chinowsky 90, Working Paper) have been studying this problem in the "Knowledge-based System for Architectural Layout Generation" project at the Center for Integrated Facility Engineering at Stanford University. In this project, they developed a prototype automated layout generation program for the domain of university facilities. They proposed the idea of storing space relationships and layout configurations in slots within an instance of an applicable frame, and representing design principles with an IFTHEN rule form.

Chinowsky pointed out the difficulty of representing layout case histories as one of the major risks in his project (Chinowsky 90, Research Proposal). The objective of my research is to study a case-based reasoning technique which has begun to attract notable attention from AI researchers, and to examine the possibility of applying the technique to the automated floor layout generation problem.

In this paper I will introduce the basic functions of case-based reasoning systems including Memory Organizing Packages (MOPs) and Thematic Organization Points (TOPs), and some case-based reasoning applications. In addition to this fundamental study of a case-based reasoning system, I will introduce a Case-Based Reasoning Environment (CBRE) developed for this research. CBRE is the development environment for the casebased reasoning application written in Allegro Common Lisp Version 1.3.2 and FrameKit Version 2.0. The reasoning functions of CBRE are based on Micro MOPs, introduced by Riesbeck and Schank (Riesbeck and Schank 89). I translated the function of a frame in Micro MOPs into FrameKit for the purpose of attaching a demon and other basic frame functions. CBRE contains a color, interactive graphical display window to show MOPs hierarchy networks and their slots, in addition to a tracing window to show MOPs created during the reasoning process. Furthermore, I will propose the idea of applying a case-based reasoning technique to the generation of a residential house layout. Since this is still a prototype version developed in CBRE, only the room size and the building area are taken into account as design constraints.

Finally, I refer to the difficulty of extracting design knowledge from a designer, and the issue of applying a case-based reasoning technique to a generic automated floor layout generation problem.