In 28th PuK Workshop ``Planen, Scheduling und Konfigurieren, Entwerfen'' (PuK 2014), 2014. Paper Slides abstract bibtex
Planning models usually do not discriminate between different possible execution orders of the actions within a plan, as long as the sequence remains executable. As the formal planning problem is an abstraction of the real world, it can very well occur that one linearization is more favorable than the other for reasons not captured by the planning model — in particular if actions are performed by a human. Post-hoc linearization of plans is thus a way to improve the quality of a plan enactment. The cost of this transformation decouples from the planning process, and it allows to incorporate knowledge that cannot be expressed within the limitations of a certain planning formalism. In this paper we discuss the idea of finding useful plan linearizations within the formalism of hybrid planning (although the basic ideas are applicable to a broader class of planning models). We propose three concrete models for plan linearization, discuss their ramifications using the application domain of automated user-assistance, and sketch out ways how to empirically validate the assumptions underlying these user-centric models.