Modeling Probabilistic Actions for Practical Decision-Theoretic Planning

Mostexisting decision-theoretic planners represent uncertainty about the state of the world with a precisely specified probability distribution over world states. This representation is not expressive enough to modelmanyinteresting classes of practical planning problems, and renders inapplicable someabstractionbaaed planning approaches. In this paper wepropose aa a remedya moregeneral world and action model with a well-founded semantics based on probability intervals. Weintroduce the concept of interval mass assigment. Unlike massassignments, which assign a probability massto each set of states, interval mass assignmentsassign a probability interval to each set of states and are moreexpressive. Interval massassignmentsare interpreted aa representing sets of probability distributions and are used in our framework to represent the uncertainty about the state of the world. Withinthis representation, wepresent a projection rule and a methodfor computinga plan's expected utility. Weco...

• Type: Conference
• Year: 1996
• Where: AIPS
• Authors: AnHai Doan
• Keywords: AIPS 1996, Artificial Intelligence, Interval Mass, Probability Distributions, Probability Interval