The manuscript proposes a novel robust methodology for the model‐based online optimization/optimal control of fed‐batch systems, which consists of two different interacting layers executed asynchronously. The first iteratively computes robust control actions online via multi‐scenario stochastic optimization while the second iteratively re‐estimates the optimal scenario map after every single/every certain number of control action/actions. The novelty of the approach is twofold: (I) the scenario map is optimally computed/updated based on probabilistic information on the process model uncertainty as well as the sensitivity of the controlled system to the uncertain parameters; and (II) the scenario set is dynamically re‐estimated, thus accounting for the effect of disturbances and changes in the operating conditions of the target process. The proposed approach is applied to a fed‐batch Williams‐Otto process and compared to an existing multi‐scenario optimization/control algorithm as well as a non‐robust optimization/control strategy to draw conclusions about which method is more effective. © 2016 American Institute of Chemical Engineers AIChE J, 62: 3264–3284, 2016