Compiling task networks into partial order planning domains

Abstract

This paper presents theoretical results and techniques for representing and managing task network goals in the framework of partial order planning. Task oriented formalisms are more expressive than partial order based formalism for problem goals in dynamical and changing domains, but they are not more powerful. We prove that it is always possible to express a task network problem in terms of an equivalent problem stated in partial order planning formalism. The task network model has been extended to describe external events (EETN), a feature not present in many planning models. The equivalence between this new model and PO formalism is also proved.

These results allow to reuse existing partial order planners as tools in order to solve task network goals. We introduce a linear cost technique of domain transformation which compiles a given task domain in an equivalent operator based domain which is then submitted to a nonlinear planner. This technique has been successfully demonstrated by the implementation of a EETN planner based on domain tranfornnations for UCPOP.