Motivation: Upper-limb motor impairment is one of the most common consequences after Stroke. Limited capability for performing reaching and grasping movements hinders the execution of most activities of daily living. Consequently, the quality lives of the affected individuals are severely compromised. Due to these facts, the recovery of the upper limb functional capabilities is currently one of the keystones of the rehabilitation therapy. Background: Researchers are developing new methods and technologies to boost the outcomes of rehabilitation therapy. A hybrid robotic system has been proposed as a promising rehabilitation technology that combines a passive device (Armeo Spring exoskeleton) to support the arm weight against gravity with a Functional Electrical Stimulation (FES) system to execute the reaching task. This system provides to patients the possibility of training specifically and intensive exercises. Objective: The main objective of this paper is to investigate the performance and robustness of a Feedback Error learning (FEL) scheme mixed with sliding mode control (SMC) to control the FES. Methods: We implemented a nonlinear model describing the muscle response to FES and the dynamic behavior of the elbow joint. Using this model we carried out a simulation study to compare four control strategies: computed torque control (CTC), sliding mode Control (SMC), and adaptive feedback control using FEL: ANN+ CTC and FEL: ANN+SMC. We tested these controllers in two different simulation conditions: In the absence and presence of fatigue. To check the performance of the controllers, we compared the root means square (RMSE) of tracking error and the Normalized RMS of muscle stimulation for various range of movement (ROM). Results: All four controllers achieved good tracking performance in the absence of perturbations. When introducing muscle fatigue, good tracking performance is given essentially by the adaptive control ANN+SMC. Conclusion: Among the proposed approaches, we conclude that the adaptive control (FEL: ANN + SMC) is the most efficient and robust controller, which has been proven by calculating RMSE.
Feedback error learning and nonlinear adaptive control