Adaptive and Neural Network-based Control Methods Comparison using different Human Torque Synthesis for Upper-limb Robotic Exoskeletons

The unprecedented and exponentially growing global senior population is creating an exorbitant and unmet demand for physical rehabilitation. Telerehabilitation with robotic exoskeletons is an emerging, and compelling complementary rehabilitation modality. Some challenges are to overcome the effects of dynamic modeling uncertainties and ensure good tracking performance, stability, safe and compliant motion, and a high degree of telepresence between the two remotely-separated human-robot systems in the presence of nonlinearities, human torques, and communication constraints such as time delays. Two control methods were developed: Adaptive Robust Integral Impedance model (ARII) control and Adaptive Robust Integral Radial Basis Function Neural Networks-based Impedance model (RBFNN-I) control. Both methods implement compliant behaviour using an adjustable impedance model and revealed desirable performance. A novel human torque regulator (HTR) was developed, which provides higher fidelity telepresence for the therapist compared to existing methods to enhance the safety and perception of the closed-loop physical interaction. Unilateral and bilateral simulations were carried out using two-degrees-of-freedom (2-DOF) exoskeletons models and experiments were performed using single-joint robots. Excellent tracking performance, telepresence, and stability was achieved in the presence of large, variable and asymmetric time delays and human torques under numerous parameters variations.