Iterative learning for robust control

This paper presents a new approach to controller design based on model unfalsification during a sequence of experiments. The general formulation of the method will allow for the use of various methods of robust control. The two most important cases of H_∞ and l_∞-norm-based robust control can both be accommodated within this general framework. One of the most important questions is whether the sequence of control designs will converge and whether the solution found will be optimal in some sense. To answer these questions, the convergence result will state that the method not only converges under mild conditions but the final controller is nearly optimal from the allowed set of model and controller structures a priori considered