An EMG-based approach for on-line predicted torque control in robotic-assisted rehabilitation

This paper proposes a sEMG-based method for on-line torque prediction and control of robot joints. More in detail, the Mean Absolute Value (MAV) features extracted from the sEMG signals acquired from five muscles of the shoulder and of the elbow are used as input to two trained time delayed neural networks (TDNNs) to estimate the joint torque of an active exoskeleton robot for movements executed in the sagittal plane. The sEMG-driven TDNNs, trained with a dataset composed by shoulder and elbow joint torque values registered in isometric conditions, allow to on-line control the exoskeleton joints for slow movements of the upper limbs. Finally, the method was tested and validated through experiments conducted on a healthy subject.