This paper describes collage, a planner that utilizes a variety of nontraditional methods of plan constmction within a partitioned or localized reasoning framework. The foundation of the collage approach is the use of action-based constraints. Such constraints represent domain and problem requirements directly in terms of actions, action relationships, and action parameter bindings rather than in terms of state-based goals and preconditions. In our experience, such constraints can provide a more natural vehicle for domain encoding than traditional STRIPS-based operator descriptors. In order to cope with the complexity and scale of realistic domains, COLLAGE also utilizes localization, a representational technique for partitioning problem requirements into subproblems. A localized search space consists of several smaller search spaces, one for each subproblem. collage allows these subproblem spaces to overlap and interact, and provides mechanisms for maintaining plan consistency and correctness. This combination of action-based reasoning with flexible localized search has yielded a powerful and efficient planning framework that is useful for challenging realistic domains.
Lansky, A. L. (2002). Localized planning with action-based constraints. Artificial Intelligence, 98(1–2), 49–136. https://doi.org/10.1016/s0004-3702(97)00067-2