Programming Hierarchical Task Networks in the Situation Calculus

Hierarchical Task Network (HTN) planning (Sacerdoti 1974) is an approach to planning where problem-specific knowledge is used to remedy the computational intractability of classical planning. This knowledge is in the form of task decomposition directives, i.e. the planner is given a set of methods that tell it how a high-level task can be decomposed into lower-level tasks. The HTN planning problem consists in computing a sequence of primitive tasks that corresponds to performing the initial set of high-level tasks. Our purpose in this paper is 1) to give an account of HTNplanning as high-level programming in the situation calculus (McCarthy 1963) based languages Golog/ConGolog (Levesque et al. 1997; De Giacomo, Lesperance, & Levesque 2000) and 2) to illustrate our approach with a ConGolog encoding of a logistics domain HTN-planning problem. The Golog/ConGolog languages have been extended to deal with explicit time, sensing actions, exogenous events, execution monitoring, incomplete knowledge of the initial state, stochastic actions and others. Thus the range of problems that can be tackled with this approach is potentially much larger. As an example, we modified the logistics domain encoding to execute on-line and deal with run-time exogenous delivery requests.