In this study, an electromyography (EMG) signal-based learning is integrated with a Sliding-Mode Control (SMC) law for an effective human-exoskeleton synergy. A modified Recursive Newton-Euler Algorithm (RNEA) with SMC was used to determine and control the inverse dynamics of a highly nonlinear 4 degree-of-freedom exoskeleton designed for the automation of upper-limp therapeutic exercises. The exoskeleton position and velocity, along with the raw EMG signal from the bicep Brachii muscle were used as a feedback. The root mean square (RMS) values of targeted muscles EMG are tracked in a predetermined time window to quantify an adaptive threshold value and control the torque at the exoskeleton joints accordingly. Simulations of the proposed robust control law have been done in MATLAB-Simulink. Results have shown that the designed hybrid Control strategy offers the ability to adjust the needed support instantly based on the amount of muscle engagement presented in the combined motion of the human-exoskeleton system while maintaining the state trajectory errors and input torque bounded to ±7 × 10?3 rads and ±5 N.m, respectively.