Sensorless Control with Friction and Human Intention Estimation of Exoskeleton Robot for Upper-limb Rehabilitation

In this paper, for the control problem of upper-limb rehabilitation using exoskeleton robot, we propose a sensorless control scheme with human intention estimation. In order to implement the active mode rehabilitation therapy, an interactive torque observer using Kalman filter is utilized to obtain human intention on the human-robot interaction control. Since the accurate friction model is crucial for constructing the observer, a deep neural network (DNN) is proposed in this study to obtain an accurate friction model. Furthermore, a variable admittance model is constructed to derive human intention to the desired motion trajectory. Various experiments have been conducted to verify the performance of the proposed control scheme based on the interactive torque observer.