Control Strategies in Planning
Froduald Kabanza · 1995
Over the years increasing sophisticated planning algorithms have been developed. These have made for more efficient planners. However, current state of the art planners still suffer from severe complexity problems, problems that can surface even in domains in which good plans are easy to generate, like the blocks world. Planners generally employ search to find plans, and planning research has identified a number of different spaces in which search can be performed. Of these, three of the most common are (1) the forward-chaining search space, (2) the backward-chaining search space, and (3) the space of partially ordered plans. The forwardchaining space is generated by applying all applicable actions to every state starting with the initial state; the backward-chaining space by regressing the goal conditions back through actions that achieve at least one of the subgoals; and the space of partially ordered plans by applying a collection of plan modification operators to an initial dummy plan. Backward-chaining and partial-order planning both have a significant advantage over forward-chaining in that they are goal directed: they never consider actions that are not syntactically relevant to the goal. Partial-order planning has an additional advantage over backward-chaining in that it explores partially ordered plans. This means that the search algorithm can detect at every point in its search space whether or not various actions interact, and impose an ordering between them only if they do. Linear backward or forward chaining planners might have to backtrack over an exponential number of improper orderings. However, both backward-chaining and partial-order planners search in spaces in which knowledge of the state of the world is incomplete. For example, computing whether or not a literal holds at a particular point in a partially-ordered plan is only tractable in certain restricted cases [Cha87]. In the forward-chaining space, on the other hand, for most planners, the points are complete world descriptions) Hence, there is not the same