On initial conditions in iterative learning control