A computational procedure for suboptimal robust controls

We consider a class of optimal control problems that depend on a set of scalar parameters which could have some uncertainty as to their exact values. We show how to compute the control functions given that we wish to balance two objectives. The first is the original objective while the second is the variation of the original objective with respect to the scalar parameters. That is we wish to move the controls to a position where there is less variation with respect to uncertainty in the scalar parameters, perhaps at the expense of the original objective. The gradient of the combined objective is derived and the method demonstrated using two examples.