On Plan Adaptation through Planning Graph Analysis

Abstract

Fast plan adaptation is important in many AI applications requiring a plan management module. From a theoretical point of view, in the worst case plan adaptation is no more efficient than a complete regeneration of the plan. However, in practice adapting an existing plan can be much more efficient than generating a new one from scratch, especially when the changes to the plan that are required concern only some circumscribed parts of the plan. In this paper we discuss a simple plan-adaptation method based on Blum and Furst’s Planning Graphs approach. The method is domain-independent and exploits the planning graph structure for a fast identification of the flaws that are present in the plan, and for fixing them by replanning limited portions of the plan. We present results from some experiments aimed at testing our method with several modifications of planning problems that are hard to solve for current planners based on planning graphs, such as IPP, Graphplan, and Blackbox. These results show that the method in practice is very efficient, especially when the plan can be adapted by changes that are localized in restricted parts of the original plan.