Extending PDDL to mondeterminism, limited sensing and iterative conditional plans

The last decade has witnessed a dramatic progress in the variety and performance of techniques and tools for classical planning. Much of this can be ascribed to the existence of a de-facto standard modeling language for classical planning, PDDL. PDDL has fostered information sharing and data exchange in the planning community, and has made international classical planning competitions possible. At the same time, in the last few years, non-classical planning has gained considerable attention, due to its capability to capture relevant features of real-life domains which the classical framework fails to express. However, no significant effort has been made to achieve a standard means for expressing non-classical problems, making it difficult for the planning community to compare non-classical approaches and systems. This paper provides a contribution in this direction. We extend PDDL in order to express three very relevant features outside classical planning: uncertainty in the initial state, nondeterministic actions, and partial observability. NPDDL's extensions are designed to retain backward compatibility with PDDL, and with an emphasis on compactness of the representation. Moreover, we define a powerful, user-friendly plan language to go together with NPDDL. The language allows expressing program-like plans with branching and iterations structures, as it is necessary to solve planning problems in the presence of initial state uncertainty, nondeterminism and partial observability. We are testing NPDDL's ability to cope with a variety of problems, as they are handled by a state-of-the-art planner, MB