DOI:
Abstract:
Two techniques to provide inferential tractability for decision-theoretic planning are discussed. Knowledge compilation is a speedup technique that pre-computes intermediate quantifies of a domain model to yield a more efficiently executable model. The basic options for compiling decision-theoretic models are defined, leading to the most compiled model in the form of parameterized, conditionaction decision rules. We describe a general compilation algorithm that can be applied to dynamic influence diagrams for sequential decision-making. Decision-theoretic raetareasoning is a control technique that regulates the online replanning. This algorithm focuses on the most promising areas to replan and considers the time cost of replanning.