Interactive Torque Controller with Electromyography Intention Prediction Implemented on Exoskeleton Robot NTUH-II

Various shoulder joint movement-related impairments, either orthopedic- or neurological-originated, clinically present characteristics of insufficient muscle strength, altered muscle groups firing patterns or inability to actively control the joint that may lead to even worse conditions such as pain, stiffness or shoulder impingement syndrome. Through continuous physical therapy and motor learning strategies, patients with shoulder disorders can gain functional improvements and reeducate the muscle contraction patterns. To retrieve a normal rhythm of muscle activation, intensive and consistent training is required, which thrives in the demand of robotics rehabilitation that aims in assisting therapists and promote cost effectiveness during prolonged rehabilitation process. In this paper, we present a novel interactive torque controller with electromyography intention prediction implemented on NTUH-II exoskeleton rehabilitation robot arm. K-nearest neighbors (KNN) algorithms are used to predict the user's intended moving direction, and the interactive torque observer is used to change the dynamic behavior of the exoskeleton robot to make it lighter and movable. As a result, we apply this method to the active control of rehabilitation and activities of daily living (ADL) tasks. To validate the effectiveness, the proposed control methodology and to illustrate the performance of the resulting robot system, we have conducted a series of experiments, and the test results are quite appealing.