• electromyography
• Gaussian mixture model
• high-level controller
• intention recognition
• muscle synergy
• upper-limb exoskeleton

Several common diseases, such as stroke, multiple sclerosis, and spinal cord injuries, lead to upper-limb impairments, affecting the ability to perform activities of daily living. Among robotic devices, upper-limb exoskeletons have been found effective in the rehabilitation of impaired subjects, helping them to improve their residual capabilities. The aim of this work is to develop an electromyography (EMG)-based control strategy for using upper-limb exoskeletons during rehabilitative exercises consisting in planar reaching tasks. In this thesis, an EMG-based high-level control algorithm for user’s movement onset and intention detection is presented, describing its real-time implementation and human-in-the-loop testing. The movement intention detection algorithm consisted of a muscle synergy-based prediction algorithm, implementing an evidence accumulation approach. In particular, the activation coefficients of muscle synergies were used as inputs of a Gaussian Mixture Model (GMM)-based classifier that estimated the reaching movement’s direction. The synergies were extracted on the real-time processed EMG-signals, acquired using a custom electromyograph. The algorithm was tested on seven healthy subjects performing planar reaching tasks towards eight targets, while wearing the NESM-gamma shoulder-elbow exoskeleton. The muscle synergies and the GMMs were subject-specific and were computed during the training phase of the algorithm. The results showed an overall accuracy, averaged over participants, of about 60%; most of the classification errors were obtained on directions adjacent to the actual one.