Abstract: To address the problem of reduced accuracy of robot motion due to the deformation error of a flexible cable during the training of an upper limb rehabilitation robot, an inverse modeling method is proposed based on the intrinsic equation of deformation of the flexible cable error. First, by combining the physiological structure and movement characteristics of the shoulder, elbow, and wrist of the human upper limb, the arm exoskeleton, which is the wearing mechanism of the rehabilitation robot, is designed to increase the compatibility of the mechanism and the human body. The robot drive system has a parallel structure of flexible cable traction, which improves the flexibility of the robot, effectively reducing the impact of robot rehabilitation training on the human upper limb, thereby making the robot perform better than that with a parallel structure. Second, based on the structural parameters and motion range of the human upper limb, the improved Denavit–Hartenberg (D–H) method was used to establish a model spatially equivalent to the wearable arm exoskeleton of a human upper limb motion in MATLAB. The rationality and correctness of the wearable arm exoskeleton mechanism design were further verified. Subsequently, the traditional method for calculating the deformation error of a flexible cable was compared with the inverse method, and the superiority of the inverse method was discussed. According to the elastic deformation mechanism in material mechanics, the deformation of the flexible cable, friction between the flexible cable and pulley, and position of the flexible cable outlet point were considered for combining the reverse solution idea with the force rotation method to calculate the deformation error of the flexible cable in the actual pulling direction. A dynamic model of the cable traction system under the expected trajectory was obtained. Based on a discussion of the motion state and quantity configuration of the flexible cable in the rehabilitation training process, a real-time dynamic model of the traction system was obtained. According to the generalized inverse matrix theory, the optimal feasible solution for the tension of each cable was determined, and the tension of the cable was obtained in real time. Finally, the relevant parameters of the system were defined, and human upper limb rehabilitation examples were provided. The deformation error of the flexible cable along the actual stress direction under the given rehabilitation training track was obtained through simulations. A comparison and analysis of the variation trend of stress and length of the flexible cable indicated that the law of deformation error of the flexible cable was consistent with the variation trend of the tension and length of the flexible cable. The length before and after the deformation of the flexible cable was compared, and the correctness of the reverse model of the flexible cable error was verified based on the flexible cable deformation equation. This provided a basis for the control and safety analysis of the flexible cable traction rehabilitation robot and a future direction for the error-solving and analysis of the flexible cable traction parallel system.