This chapter addresses the situation where there is no available mathematical model that governs the behavior of the nonlinear systems to be controlled. Nevertheless, it is possible to gather time-series representing the evolution of the measured quantities either using open-loop control (open-loop stable systems) or by operating the system under a loosely defined sub-optimal feedback control law. The chapter proposes a control design feedback that is based on the identification of the cost function as a map depending on the past measurements of the input and output on one hand and on the future control sequence on the other hand. A specific dedicated solver, based on a modified version of the Torczon algorithm, is used to address the optimizationOptimization of the possibly non continuous ML map representing the cost function. The whole framework is illustrated on a simple example and the impact of the different choices are shown.