Saddle point conditions for a class of stochastic dynamical games with imperfect information

Nonlinear dynamical games are considered in which one player has full (causal) information on the state and the opponent's measurements and control, while the other one only has partial, possibly noise-corrupted, state information. A dynamic-programming-like algorithm in an extended space is given that, when successful, yields, a saddle point, usually in mixed strategies. This result is used to provide an equilibrium in safe strategies for the case in which both players are restricted to imperfect state information.<
>