Introduction to the Special Issue on Planning: Research Issues at the Intersection of Planning and Constraint Programming

Members of the constraint programming community are usually more familiar with applications like scheduling than with action planning — the topic of this special issue. However, action planning tackles problems belonging to a similar domain and can be interpreted as an extension of the scheduling problem by the problem of selecting which actions/tasks a plan should include so that a set of given goals is satisfied/optimized. The connections between planning and constraint satisfaction have been recognized, at least since the development of the MOLGEN planner (Stefik, 1981). Much of the subsequent work at the intersection of planning and CSP involved tackling subproblems of plan synthesis that can be posed as constraint satisfaction problems, such as the management of resources, temporal constraints, and the domains of planning objects. Well-known planning systems like SIPE (Wilkins, 1988), OPlan (Tate et al., 1994) and IxTeT (Laborie and Ghallab, 1995) have posed such tasks as constraint satisfaction problems. The planning field often benefits here from techniques developed in the context of scheduling. The advent of Graphplan (Blum and Furst, 1997) further strengthened the connection between plan synthesis and constraint programming. Many of the explanations of Graphplan’s and its extensions’ impressive performance pointed to the connections between its graph search and CSP techniques (Kambhampati, 2000; Rintanen, 1998). Work on variants of Graphplan has shown that bounded-length plan finding (i.e., finding if a plan of length k exists for a given problem) can be usefully posed as a model finding problem, and solved using CSP, SAT or IP techniques (Kautz and Selman, 1996; Do and Kambhampati, 2001; Kautz and Selman, 1999; Bockmayr and Dimopoulos, 1999; Vossen et. al., 1999; Kautz and Walser, 1999).