Leveraging second-order information for tuning of inverse optimal controllers