Optimal control of Markov decision processes with incomplete state estimation