Numerous researches have been carried out to use robot in the area of stroke rehabilitations. The conventional approach is to assist patient to perform Activities of Daily Living (ADL) through a set of pre-programmed trajectories. The main drawback lies in the lack of voluntary movement by the patient. Many of these works have been carried out by using positional feedback control. In this kind of system, it is difficult to assess patient's muscle condition, which may results in inconveniences due to torques or forces asserted by the rehabilitation robot. The main aim of the project is to build a stroke rehabilitation system using socially inspired robot technique. The system monitors patient's muscle activity and uses this information to drive an exoskeleton robot that will assist patient to his/her arm. Movement is generated based on voluntary muscle activity by patient and therefore will improve their learning curve. Another main advantage is the system minimizes patient's inconveniences due to movement by robot. The system is based on torque feedback control. A sliding mode control was implemented in replace of conventional control. The main advantage of sliding mode control over conventional control is its robustness. Sliding mode control does not require precise mathematical model of the system and it is insensitive to parametric changes and uncertainties within the system.