We describe a framework for designing controllers for general, partially
observed systems which are robust w.r.t. uncertainties; both parametric and
structural. The framework is founded on stochastic decision making by employing
maximum entropy stochastic models for the nonlinear system. The framework
emphasizes the significance of the Lagrange multipliers involved in the maximum
entropy model construction. They provide sensitivities of controller
performance
vs. uncertainty model complexity - relative value of a model for a control
objective. This relationship is further established via duality theory. We then
develop a similar framework for networked systems and describe a result which
attempts to quantify the duality between information patterns and multi-agent
control performance.