In the current robot aided rehabilitation training methods based on patient’s active participation, the perception of the patients’ active participation are mostly from the viewpoint of “motor” while neglecting the “psychological” involvement. Moreover, the interactive control methods do not incorporate robotic continuous variable motion control and therapist discrete event decision control into a unified framework. To solve this problem, a robotassisted clinical rehabilitation training method is proposed based on anxiety emotion recognition and hybrid theory. Firstly, the significance and difference of anxiety with different arousal are analyzed; Secondly, an anxiety emotion classifier using radial basis function based support vector machine is developed. Besides, humanrobot interactive controller coinciding with the arousal of the patient’s anxiety is developed by using hybrid control theory. Finally, the proposed method is verified with three recruited stroke patients on a constructed clinical experimental platform featuring a Barrett WAMTM manipulator.