Operator Selection While Planning Under Uncertainty
This addresses planning in uncertain environments for AI systems, but it appears incremental as it builds on existing methods like Dempster-Shafer theory and abstraction hierarchies.
The paper tackles the problem of planning under uncertainty by constructing quantitatively ranked plans using a best-first search strategy guided by expected fulfillment, resulting in the generation of multiple plans and a super-plan with knowledge acquisition operators for decision-making during execution.
This paper describes the best first search strategy used by U-Plan (Mansell 1993a), a planning system that constructs quantitatively ranked plans given an incomplete description of an uncertain environment. U-Plan uses uncertain and incomplete evidence de scribing the environment, characterizes it using a Dempster-Shafer interval, and generates a set of possible world states. Plan construction takes place in an abstraction hierarchy where strategic decisions are made before tactical decisions. Search through this abstraction hierarchy is guided by a quantitative measure (expected fulfillment) based on decision theory. The search strategy is best first with the provision to update expected fulfillment and review previous decisions in the light of planning developments. U-Plan generates multiple plans for multiple possible worlds, and attempts to use existing plans for new world situations. A super-plan is then constructed, based on merging the set of plans and appropriately timed knowledge acquisition operators, which are used to decide between plan alternatives during plan execution.