Robust Control of Markov Decision Processes and Connection to Risk-sensitive Control

@inproceedings{Coraluppi1996RobustCO,
  title={Robust Control of Markov Decision Processes and Connection to Risk-sensitive Control},
  author={Stefano P. Coraluppi and Steven Marcus},
  year={1996}
}
This paper introduces a formulation of the robust control problem in the partially observed Markov Decision Process (POMDP) setting. We show that this formulation is the large risk limit of the risk-sensitive control problem. Exploiting this connection, we derive an information state process and dynamic programming equations for the value function. We develop a methodology to determine an optimal policy for nite horizon problems, and near-optimal policies on the innnite horizon. Finally, we… CONTINUE READING