“…To attain high tracking accuracy under model uncertainties at run-time, e.g., due to manipulation of an unknown object, it is common to use model-based adaptive control, which performs online parameter adaptation driven by tracking errors [1], prediction errors [2], or both error sources in a composite manner [3]. Recently, Lee et al [4] proposed a natural adaptation law that guarantees the physical consistency [5], [6] of the estimated inertial parameters, with additional advantages of improved transient performance and less laborious gain selection; a recent application can be found in [7]. However, these adaptive control schemes only adapt the feedforward term but do not adjust the feedback term according to model uncertainties.…”