DOI:
Abstract:
Planning variables are fundamental to all modern nonlinear planners. Deriving plans for complex domains typically means managing many variables for a single plan. Since planning is concerned with reasoning about actions over time, determining the possible values of these variables at the various time points represented by a plan while the plan is being constructed is no small task. Backtracking over certain choices made during plan construction, thereby changing some variable values, further complicates the management of variables. This paper discusses issues in representing and applying constraints on planning variables with regard to these complexities. Specifically, it presents a method to assert constraints on planning variables using a justification-based scheme that allows nonmonotonic retraction of variable codesignations. This system of managing defeasible constraints on variable values allows the planner to rely on the codesignation system to help prune the search space of possible plans without compromising the least-commitment approach desired in domain-independent planners.