Learning feed-forward control by