Probabilistic Planning with Nonlinear Utility Functions

All Papers

Researchers often express probabilistic planning problems as Markov decision process models and then maximize the expected total reward. However, it is often rational to maximize the expected utility of the total reward for a given nonlinear utility function, for example, to model attitudes towards risk in high-stake decision situations. In this paper, we give an overview of basic techniques for probabilistic planning with nonlinear utility functions, including functional value iteration and a backward induction method for one-switch utility functions.