RBF Neural Network-Sliding Model Control Approach for Lower Limb Rehabilitation Robot Based on Gait Trajectories of SEMG Estimation

This paper designed and developed a new RBF neural network-sliding model controller for patients with stroke and lower extremity motor dysfunction, and applied it to a 3 degrees of freedom (3-DOF) lower limb rehabilitation robot (LLRR) for passive rehabilitation of patients. At first, a simple LLRR structure is designed that can be adjusted to fit the patient at the hip, knee, and ankle joints. Then, the patient's sEMG signal is obtained to predict the expected trajectory of the LLRR system, where the EMG signal is detected by BIOPAC software. Moreover, a RBF neural network-sliding model approach is designed for the dynamics model of the LLRR, and the asymptotic stability of the controller is verified via a Lyapunov theorem. Finally, LLRR system is experimentally verified by the MATLAB software, which exploit that the proposed control approach is feasible and effective for the lower extremity patients. Thereby, the developed control approach has illustrated high efficiency and robustness for the patient's passive rehabilitation training in real-time.