Command and Control (C 2) decisions must of necessity be based on imperfect knowledge of the battlespace. With the advent of unmanned air vehicles (UAVs), the pace at which decision points arise will increase. This necessitates the development of automated C 2 tools for use in the decision-making process. Estimation and control in the presence of purely random input noise is well-understood, and produces excellent results. In the context of C 2 decisions, one most consider observations contaminated by both random noise and adversarially induced "noise". Consequently, zero-sum, discrete, stochastic games under imperfect observations are considered here. Machinery has recently been developed which allows one to solve such problems. The theory is summarized. For problems in the class considered here, the resulting algorithms are computationally feasible. The method is applied on a small game testbed. The behavior of the resulting controls are discussed. An alternate (naive) approach is to apply the optimal state feedback game controls to the maximum likelihood state. This alternate approach is susceptible to deception by the opponent. It is shown that the improvements in using the robust approach range from small to tremendous depending on certain factors.