Extending Artificial Intelligence Techniques for Hierarchical Planning

Generating a plan of correctly sequenced activities to achieve a stated goal under given constraints is a difficult task for either humans or computers. For any realistic planning problem, there are almost infinite combinations of activities that can be generated, of which few—or, in some cases, none—will satisfy the constraints applicable to the problem. Automated planning has been a Holy Grail of AI research since the 1960s. In spite of this, little progress has been made in generating plans except for all but the most trivial block‐stacking problems until recently. This paper explains how SIPE‐2 (System for Interactive Planning and Execution) was used to generate hierarchical plans for the construction of a single‐family house, culminating in a set of detailed plans for each trade that are then integrated and reconciled with one another. The use of SIPE‐2 for the house‐construction problem was one of a number of experiments carried out by the writers to address the feasibility of AI planners in construction. The representation and reasoning used by SIPE‐2 to perform hierarchical planning of the house construction are the focus of this paper. Papers dealing with other aspects of the research are cited throughout.