The STRIPS Assumption for Planning Under Uncertainty

Abstract

The virtue of the STRIPS for planning is that it bounds the information relevant to determining the effects of actions. Viewing the assumption as a statement about beliefs, we find that it does not actually assume anything about the world itself. We can characterize the assertion about beliefs in terms of probabilistic independence, thereby facilitating analysis of representations for planning under uncertainty. This interpretation separates the STRIPS from other necessary features of a planning architecture, such as its model of persistence and its inferential policies. By isolating these factors, we can understand the role of dependence across a wide range of planners and action representations. Graphical models of dependence developed for probabilistic analysis provide a convenient tool for verifying the STRIPS for a variety of planning systems. Investigation of a few representative systems reveals a Markovian event structure common to these planning models.