Real-time Multiple-Channel Shoulder EMG Processing for a Rehabilitative Upper-limb Exoskeleton Motion Control Using ANN Machine Learning

Myoelectric Control Systems for Upper Limb Wearable Robotic Exoskeletons and Exosuits—A Systematic Review

“…Therefore, most of the exoskeletons supporting motions of two or multiple joints, such as shoulder–elbow [ 29 , 47 ], elbow–wrist [ 8 , 26 ], or shoulder–elbow–wrist [ 16 , 17 ], found in the literature, were not portable. Apart from that, some studies also mentioned fixed exoskeletons supporting a single joint motion, which were mainly elbow exoskeletons [ 9 , 17 , 21 , 25 , 30 , 33 , 34 , 37 , 40 , 47 , 53 , 62 , 63 ] (except one shoulder exoskeleton in [ 68 ] and one hand exoskeleton in [ 36 ]), which were not portable because of their high weight-to-power ratio.…”

“…Rehabilitative upper limb exoskeletons can aid patients with motor disabilities in performing therapeutic exercises, in order to restore/improve the motor functions in their impaired limbs. A total of 58% (34 articles) of the exoskeletons in the reviewed papers were intended for rehabilitation scenarios, assisting the patient in performing different single or multiple joint motions (some examples include [ 15 , 16 , 21 , 24 , 29 , 68 ]). The benefits of wearable robot-assisted rehabilitation compared to traditional physical therapy include the provision of repetitive and intensive training to the patient, lessening the physical burden of the therapist, and offering an effective means of objectively quantifying the patient’s progress before and after training [ 1 ].…”

“…In addition, a few studies utilized isometric contraction tests to evaluate the effect of the developed prototype on muscle fatigue [ 23 , 64 ]. Classification accuracy and confusion matrix outcomes commonly evaluated the ML-based myoelectric control systems’ performances and computation times (or latencies) [ 18 , 63 , 64 , 67 , 68 , 69 ]. Only one study evaluated the learning time for a model-based reinforcement learning framework, and found that the proposed method could learn the proper assistive strategy after only 60 seconds of user–robot interaction [ 40 ].…”