Planning is a ubiquitous aspect of the behavior of intelligent agents. And, it is crucially embedded in our intuitive ideas of rationality. A rational agent creates a plan to achieve a goal which is consistent with the agent's knowledge of the world and of the affects that its actions will have on that world. This is the ideal, but as we will see this ideal can rarely be met. Further , we must plan in widely varying circumstances. Sometimes, our knowledge of the world in which we must act is severely limited or so uncertain as to preclude the possibility of guaranteeing the success of our plan no matter how carefully crafted. Often our plans must of necessity involve a good deal of "real time" improvisation or large scale revision. And, even the best of plans will often fail.

     This section begins with one of the earliest search methods, means-end analysis. In contrast with state space search methods, this procedure can be used as a generator of subproblems (or subgoals) and this characteristic is central to much of the work on planning. A complementary focus is on problems of representation. In linear planning subgoals are generated by determining whether or not they are satisfied in the current representation of the state of the planning world. In non-linear planning subgoals are generated by determining whether or not they are represented within the current plan representation.

     The description of the work begins with research carried out within the "Standard" AI Planning Paradigm. Next we turn to some work that relaxes the assumptions of this standard planning paradigm. Finally, we end with a brief consideration of the problem of Plan Recognition. Plan Recognition can be view as a kind of dual of planning. However, it also surfaces as a component of a planning process if we consider planning in the context of other intelligent agents.