Enabling-Condition Interactions and Finding Good Plans*
Dana S. Nau · 1993
In AI planning research, the best-known goal interaction is the deleted-condition interaction, in which the side-effect of achieving one condition is to delete some other condition that will be needed later. Enablingcondition interactions—in which the side-effect of achieving one condition is to make it easier to achieve some other condition—are not as well known. In this paper, I argue that enabling-condition interactions merit more attention than they have heretofore received. This paper is organized as follows: 1. A definition of enabling-condition interactions. 2. The importance of finding “good ” plans (rather than simply being satisfied with any plan we find), with examples from several planning domains. 3. How a number of planning strategies take advantage of enabling-condition interactions to produce better plans. 4. Some of the effects of enabling-condition interactions on the complexity of planning. 5. Concluding remarks. Definition An enabling-condition interaction is a situation in which some action invoked to achieve one goal G1 also makes it easier to achieve another goal G2. For example, in Figure 1, the action move(a, c, b) achieves the goal on(a, b), but it also has the side-effect of clearing c, making it easier to achieve the goal on(c, d). As another example, consider the following situation (based on (Wilensky, 1983)): John lives two miles from a bakery and two miles from a dairy. The two stores are one mile apart. John has two goals: to buy bread and to buy milk.