A Framework for Hybrid Planning

Hybrid models are characterised by the integration of two (or more) different paradigms of representation within the same system. Most current planning problem description languages are purely sentential, i.e., based on predicate logic formalisms [3, 14]. The ramification of the effects of action [16] makes the sentential description of planning problems that involve the spatial movement of physical entities over-complex and inefficient. Evidence from research in knowledge representation and planning [6, 9, 8] indicates that these problems are more effectively modelled using homomorphic [1] (or analogical) representations. As for sentential descriptions, however, the complexity of the real world prevents purely homomorphic formalisms from being always the most effective choice. This paper proposes a model-based theoretical framework for planning with hybrid representations, in which equivalent sentential and analogical models can be simultaneously and interchangeably used. The analogical model which is developed extends a recently proposed representation for purely homomorphic planning [5]. The sentential model of action adopted is based on the current standard planning domain description language PDDL2.1 [3]. The result is a powerful, heterogeneous planning representation that overcomes the limitations and offers the complementary strenghts of the two formalisms on which it relies.