Least Commitment { an Optimal Planning Strategy

We develop an extension of Least Commitment Search (LCS), which represents a new strategy for recovering from failure in planning. The particular type of failure addressed here is that which requires extra actions to be introduced into the plan. The strategy is explored in the context of an optimisation problem, that of generating plans of minimal length when given a single resource. There are two main results in this paper, both of which are established empirically. The rst is that LCS is eeective as a strategy for introducing new actions. The second is that it is a limited strategy. When the planner is required to introduce a substantial number of new actions, i.e. when search becomes relatively deep, LCS is not in itself an adequate strategy. The challenge which emerges is one which planners have largely ignored. Is there an eeective general strategy for dealing with the task of introducing extra actions? This is a question not only for planners which seek to optimise on certain dimensions; it is also a problem for planners which seek only to establish feasibility.