Experimental dynamic identification and model feed-forward control of Novint Falcon haptic device

Abstract This paper addresses the dynamic identification and model feed-forward control implementation of a Novint Falcon haptic device. Firstly, inverse and forward kinematic problem of the device is obtained. Subsequently, equations describing the robot’s dynamics is derived using Newton–Euler approach which includes friction of active joints. The unknown dynamic parameters including mass and moment of inertia of constituting parts along with dry and viscous friction coefficients is identified practically by a white-box identification approach using LS technique. Furthermore, the torque digital scale coefficient which Novint Falcon’s firmware uses is identified. To the end of obtaining some of regressors a novel numerical-based differentiation approach is presented. Having an experimentally verified dynamic model in hand, the model is used in a model feed-forward impedance controller in order to omit the effect of natural dynamics of the robot, including active joints frictions from displayed impedance. Experimental tests reveal that the implemented force sensor-less controller performs 26% more accurate in terms of output force comparing device’s official controller.