Risk-Sensitive Optimal Control of Partially Observable Markov Chains
Risk-Sensitive Optimal Control of Partially Observable Markov Chains Emmanuel Fernandez-Gaucherand, Steven I. Marcus --- U of AZ, SIE Dept., Tucson, AZ 85721 ,
We study the finite-horizon optimal control of partially observable Markov chains under a risk-sensitive criterion. In particular, an exponential utility function is used. General results on sufficient statistics, dynamic programming equations and structural results of value functions are obtained. Further, a case study considered in detail reveals qualitative properties of risk-sensitive policies.