Context dependent effects in temporal planning

Abstract

A major problem in planning, as in most AI domains, is to find an adequate representation. In particular, there is the issue of which effects of an action should be specified unconditionally in its model and which can be stated conditionally with respect to the context. Indeed, most actions do have several different effects depending on the context in which they are executed. In this paper, we propose an approach and different extensions in order to take into account context dependent effects into IXTET, a temporal planner. Expressiveness requires a great flexibility of representation, but it may lead to a computational cost not compatible with a practically efficient planner. The proposed approach offers a slight extension in representation which enables to express conditional subtasks. Furthermore, empirical results show that this approach and the corresponding implementation provide also some efficiency benefits with respect to a domain description that details all unconditional action models.