Functional electrical stimulation (FES) is a potential technique to provide active improvement to spinal cord injured (SCI) patients in terms of mobility, stability and side-effect prevention. FES-elicited muscle force is required to be appropriate and persistent to perform intended movement or maintain a posture balance. However, muscle state changes such as muscle fatigue degrade the performance of FES. In addition, most of complete SCI patients don't have sensory feedback to detect the fatigue and in-vivo joint torque sensor is not available yet. Conventional FES control systems are either in open-loop or not robust to muscle state changes. Therefore, this thesis aims at a development of joint torque prediction and feedback control method in order to enhance the joint torque control of FES in terms of accuracy, robustness, and safety to the patients. In order to predict FES-induced joint torque, evoked-Electromyography (eEMG) has been applied to correlate the muscle electrical activity and mechanical activity. Although muscle fatigue represents time-variant, subjectspecific and protocol-specific characteristics, the proposed Kalman filter-based adaptive identification is able to predict the torque generation systematically. The robustness of the torque prediction has been investigated in a fatigue tracking task through experiments in SCI subjects. The results demonstrate good tracking performance against muscle state variations and external disturbances. Based on accurate predictive performance of the proposed method, a new control strategy, EMG-Feedback Predictive Control (EFPC), is proposed to adaptively adjust stimulation pattern to obtain the desired joint torque trajectory. This control strategy is not only able to explicitly avoid over-stimulation to the patients and conveniently generate appropriate stimulation pattern for the desired torque trajectory, but also able to track the desired torque trajectory compensating time-variant muscle state changes.
