Partially observed semi-Markov zero-sum games with average payoff (Q930973)

The article analyzes a two person zero-sum stochastic dynamic game on a countable state space, with average payoff and partial observations. It is shown that one may consider an associated completely observed game whose state for each agent is the corresponding recursively computable conditional law of the state given past actions and the agent's past observations, insofar as a saddle point equilibrium for this corresponds to one for the original game. Using this, the existence of a saddle point equilibrium is established. The well-posedness of the associated Shapley equation is established and a value iteration algorithm is given for the solution thereof.