A Constraint-Programmed Planner for Deep Space Exploration Problems With Table Constraints

Abstract

In recent years because of the increasing number and types of scientific payloads on a probe, the constraints between payload and probe and the constraints among payloads have become increasingly complex. The technology of constraint processing has gradually become a focus of research in deep space planning. In this paper, we propose a constrained-programmed planner called DSPlan for deep space planning problems based on table constraints. We first propose a technique for automatically converting the planning domain definition language model used in planning into the form of table constraints. Following the common practice of coding a planning problem as a constraint satisfaction problem with multiple levels, we propose a dynamic constraint set and a corresponding mutex filtering algorithm to express the different combinations of constraints on varying levels that need to be satisfied. This new form of data structure uses explicit domain information to maintain the generalized arc consistency of table constraints. Empirical analyses demonstrate the efficiency of table constraints over the international planning competition problem and classic deep space instances in general arc consistency schema algorithms. Experimental results also prove that DSPlan and table constraints are highly promising general-purpose tools for deep space planning problems compared with other planners.