SightPlan Experiments: Alternate Strategies for Site Layout Design

Abstract

SightPlan is a knowledge-based system that lays out temporary facilities on construction sites. Its implementation architecture calls for explicit representation of the problem-solving strategy used to construct a solution layout. Because this strategy can easily be modified we explore several alternate formulations. This paper describes the development and testing of strategies that two agents might adopt to match their cognitive capabilities. One strategy models the steps taken by a construction manager laying out the site; it reflects human cognitive strengths and weaknesses. Recognizing that such a strategy would not be well-suited for execution by a computer, we critique it, and craft a second strategy to make better use of the computer’s storage and computational capabilities. Although more powerful in several ways, this second strategy falls short of some human strengths, such as the ability to discriminate between alternatives. After comparing these two strategies, we develop a third, combining human and machine strengths. This strategy augments the second strategy with graphics to allow for human-computer interaction. We argue that the resulting joint cognitive system can generate better solutions than either a person or a machine.