All Published Papers Search Service

Title

GILAO*: Solving Uncertainty Planning with Comparison of Generalized Intervals

Author

Jigui Sun, Shuai Lu, Minghao Yin, Shulin Cui

Vol. 6  No. 1  pp. 57~63

Abstract

Classical uncertainty planning problems assume that the probabilistic models of the domain are always accurate. Unfortunately, the existed uncertainty planning algorithms couldn’t be used to solve the problems with incomplete knowledge, and then give an optimal plan or a less optimal plan. This paper proposes a method, named Comparison of Generalized Intervals. it selects the average model form different possible models by a heuristic function. Furthermore, this paper gives a new algorithm, named GILAO*, to solve the uncertainty planning problems, in which each successor state could be assigned a transitional probabilistic interval. Because GILAO* considers not only the worst or best model, but also all other possible models, it is more generalized than the robust algorithms. The possible action is selected by CGLs so that algorithm GILAO* is efficient to solve uncertainty planning problems. The experimental results show this. Moreover, GILAO* inherit the merit of LAO* algorithm, namely it need not expand the whole state spaces to find the optimal plan.

Keywords

uncertainty planning, comparison of generalized intervals, Markov Decision Process

URL