Planning in design synthesis : abstraction-based LOOS (ABLOOS)

TLDR: This proposal describes the creation of a new frameworic to represent and control decompositional abstraction levels in layout design that will not only help to tame the explosion in the search space but will also add flexibility and adaptability to the approach.

Abstract: ion is the quintessential mechanism for dealing with complexity. ion may take many forms and appears to underlie all of the important strategies for handling large search spaces in planning domains. Design, modeled as search, also depends heavily on the use of mechanisms of abstraction, and may benefit from the employment of strategies to leverage search that are similar to those developed in automated planning. In some design domains, such as automated layout, the fact that the objects manipulated have shape and take up space adds an interesting twist and certain complexities to the use of abstraction mechanisms for handling the operators, objects and goals of the problem space.ion is the quintessential mechanism for dealing with complexity. ion may take many forms and appears to underlie all of the important strategies for handling large search spaces in planning domains. Design, modeled as search, also depends heavily on the use of mechanisms of abstraction, and may benefit from the employment of strategies to leverage search that are similar to those developed in automated planning. In some design domains, such as automated layout, the fact that the objects manipulated have shape and take up space adds an interesting twist and certain complexities to the use of abstraction mechanisms for handling the operators, objects and goals of the problem space. With respect to leveraging search in a design system, this proposal describes a salient topic and the research aimed at exploring it. The topic is the combination and interaction of various forms of abstraction in the context of the automated synthesis of layouts. This research issue converges with a unique opportunity to explore it in the context of an especially promising approach to space planning. The system embodying that approach, LOOS, has a representation and an overall system architecture which suitably address some of the problematic limitations and compromises of former approaches. However, in order to extend the generality and scalability of the approach, the problem of the combinatorial explosion in the search space has to be addressed. In this regard, a promising direction appears to be combining decompositional abstractions formed around the objects of design, with the primary representational abstractions used for delineation and generation of alternatives. The remainder of the proposal describes the creation of a new frameworic to represent and control decompositional abstraction levels in layout design. This new abstraction mechanism will strongly restrict the number of the objects that the layout system has to deal with at a given time, while also providing an appropriate place to store additional knowledge and constraints for pruning. Therefore, it will not only help to tame the explosion in the search space but will also add flexibility and adaptability to the approach. This new abstraction framework is proposed as a general hierarchical extension to LOOS's domain-independent problem solving structure, accepting domain-dependent decompositional knowledge as content. University Libraries Carnegie Mellon Univf Pittsburgh, Pennsylvama Table of