The proposed human-robot interface is expected to be used in the upper-limb power-assist exoskeletons for helping the daily living activities of the weak elderly or patients suffering from stroke. Generally, these people move their arms very slowly, thus, the IRD changing speed is assumed to be almost a constant during a curve motion. On the other hand, there is also a great need for assisting normal people to lift or move heavy weights while wearing an exoskeleton. In this case, the wearer may move their arms with a relatively high speed, which means that Assumption 4 does not hold. To relax this assumption, a higher dimensional Kalman filter can be designed to estimate both the speed and the acceleration of IRD. Therefore, the IRD can also be online inferred through a similar VD Kalman filter used in this study. Note that the VD filter is originally proposed for complex motion tracking and has achieved many successful applications