Systematic Nonlinear Planning

David A. McAllester, David Rosenblitt

Abstract: The field of automated planning seeks to develop fast methods for generating programs of action that allow an agent to achieve goals or maximize rewards. Initially, researchers made many simplifying assumptions, creating the classical planning problem: produce a sequence of atomic actions which will achieve a logically specified goal in a completely known world without uncertainty. McAllester and Rosenblitt’s (M&R) paper “Systematic Nonlinear Planning,” presented nineteen years ago at AAAI-91, had two major impacts on the field: 1) an elegant algorithm and 2) endorsement of the lifting technique. The paper’s biggest impact stems from its extremely clear and simple presentation of a sound and complete algorithm (known as SNLP or POP) for classical planning. While it is easy to define such an algorithm as search through the space of world states, SNLP is a “partial-order” planner, meaning it searches the space of partially-specified plans, where only partial constraints on action arguments and ordering decisions are maintained. Here, M&R benefited from Chapman’s elegant TWEAK planner (IJCAI-85), which greatly clarified previous partial-order algorithms. SNLP’s key feature is the use of a data structure, called a causal link, to record the planner’s commitment to establish a precondition of one action with the postcondition of another. While the notion of causal link wasn’t novel, originating much earlier in Tate’s NONLIN planner (IJCAI-77), SNLP was so much simpler than NONLIN that M&R could prove it was sound and complete. A bonanza of research results followed. Penberthy & Weld’s UCPOP planner (KR-92) handled actions modeled in an expressive language and was widely distributed. Other extensions included temporal planning, planning with metric resources, contingent planning in response to uncertain effects, methods for guiding the planner’s non-deterministic search, etc.