Approaches to the reuse of plan schemata in planning formalisms

Abstract

Planning in complex domains is normally a resource and time consuming process when it is purely based on first principles. Once a plan is generated it represents problem solving knowledge. It implicitly describes knowledge used by the planning system to achieve a given goal state from a particular initial state. In classical planning systems, this knowledge is often lost after the plan has been successfully executed. If such a planner has to solve the same problem again, it will spend the same planning effort to solve it and is not capable of "learning'; from its "experience';. Therefore it seems to be useful to save generated plans for a later reuse and thus, extending the problem solving knowledge possessed by the planner. The planning knowledge can now be applied to find out whether a problem can be solved by adapting an already existing plan. The aim of this paper is to analyze the problem of plan reuse and to describe the state of the art based on a variety of approaches which might contribute to a solution of the problem. It describes the main problems and results that could be of some relevance for the integration of plan reuse into a deductive planning formalism. As a result, this description of the state of the art leads to a deeper insight into the complex problem of plan reuse, but also shows that the problem itself is still far from being solved.