Cassandra is a planning system based firmly in the classical planning paradigm. Many of its strengths and weaknesses are those of other classical planning systems. For example, we believe that under certain circumstances its plans will be valid and that it is guaranteed to find a valid plan if one exists. However, the techniques it uses are valid only in limited circumstances, and its computational complexity is such as to make direct scaling up unlikely to be feasible.
In our view, the principal strengths of Cassandra arise from the explicit representation of decisions in its plans. We have shown how this use of decisions provides a natural account of how knowledge goals arise during the planning process. We have also sketched how decisions can be used as the basis of extensions that provide added functionality. A new type of decision allows fail-safe plans, which can provide a method of solving problems such as the bomb-in-the-toilet problem (Section 6.5.5); and another type of decision may provide an effective method of interleaving planning and execution (Section 7.4).
We believe that the use of explicit decision procedures will enable the extension of the range of applicability of techniques of classical planning. In general, the idea of constructing a single plan that will succeed in all circumstances is, we feel, unlikely to be productive: the real world is complex and uncertain enough that trying to predict its behavior in detail is simply impossible. However, the use of decision procedures that, for example, involve probabilistic techniques or interleave planning and execution, appears likely to provide a flexible framework that, although inevitably sacrificing completeness and correctness, will provide a basis for effective, practical planning in the real world.