Abstract: With the aim of assisting the patients with upper limb motion impairment problem to perform different types of rehabilitation training, a fuzzy sliding mode admittance control strategy is developed to realize human-robot coordinated control during training process based on the upper limb rehabilitation exoskeleton robot system. Firstly, the overall structure of rehabilitation exoskeleton and the real-time control platform are introduced. Then, the derivation process of the fuzzy sliding mode admittance control algorithm is analyzed, and the system stability is demonstrated by Lyapunov stability criterion. Finally, the passive training mode experiment and active training mode experiment are carried out under different system admittance parameters. The changing characteristics of motion deviation, human-robot interaction force, and the surface electromyography signal of bicipital muscle during experiment process are analyzed and compared. The experimental results show that the intensity and difficulty of rehabilitation training can be optimized, the human-robot interactive compliance and the active participation of patient can be improved, and the requirements of the patients with different paralysis degrees and recovery process can be satisfied by selecting appropriate objective admittance model.