Abstract: For the purpose of assisting stroke patients to perform upper limb rehabilitation training, a gravity balanced rehabilitation exoskeleton robot system is presented. Firstly, the mechanical structure of the exoskeleton and the Matlab/RTW-based semi-physical real-time control system are described. Then, the gravity balance model of the entire system is established based on the auxiliary balance method. Several zero-free-length springs and auxiliary links are utilized to balance the gravities acting upon the exoskeleton and the human arm during rehabilitation training. Finally, simulations and experiments are carried out to compare the driving torque of each joint and the surface electromyogram signal (sEMG) activities of biceps under different balance conditions. In the simulations, the average driving torque under the balance condition is about 14.89% of that under unbalance condition. The sEMG activities under the balance condition are about 57.61% and 63.49% of those under the unbalance condition while conducting two different experimental tasks. The results demonstrate the effectiveness of gravity balance in reducing driving torques and energy consumption, as well as the performance requirement of actuators.