Abstract: In the current robot-aided rehabilitation training, there still exist a certain limitations in the interaction control between the patient and the rehabilitation robot, which mainly focuses on the perception of the patients' active motor participation while not taking the patients' active psychological participation into consideration. To solve this problem, a control method for robot-aided active rehabilitation training based on emotion perception is proposed by using frustration, excitement and boredom of patients in rehabilitation training as target emotion states. Firstly, psychological responses of patients to target emotion states in robot-aided active rehabilitation training and their performance features are extracted. Secondly, a target emotion classifier using radial basis function based support vector machine is designed, and an adaptive control method for robot-aided rehabilitation tasks is developed, which can adapt to the target emotion changes of patients. Finally, the experimental system platform is constructed with the 4-DOF (degree of freedom) WAM^TM rehabilitation manipulator of Barrett Inc., and 12 stroke subjects are recruited to verify the effectiveness of the proposed method.