For stroke patients with upper limb motor dysfunction, rehabilitation training with the help of rehabilitation robots is a social development trend. Existing upper limb rehabilitation robots have difficulty fully fitting the complex motion of the human shoulder joint and have poor human–robot compatibility. In this paper, based on the anatomical structure of the human upper limb, an equivalent mechanism model of the human upper limb is established. The configuration synthesis of the upper limb rehabilitation mechanism was carried out, a variety of shoulder joint man–machine closed-chain Θ_s and shoulder elbow human–machine closed-chain Θ_se configuration combinations were synthesized, and the configuration model with compatibility and reduced moment conduction attenuation was selected from them. Two configurations, 2P_a1P3R_a and 5R_a1P, are proposed for the generalized shoulder joint mechanism of the robot. The closed-chain kinematic models of the two configurations are established, and the velocity Jacobian matrix is obtained. Motion performance analysis, condition reciprocal analysis and operability ellipsoid analysis of different configuration design schemes were carried out in different operation planes. The results show that in the normal upper limb posture of the human body, the 5R_a1P configuration of the shoulder joint has better kinematic performance. Finally, on this basis, an upper limb rehabilitation robot prototype with good human–computer compatibility is developed, and its moving space was verified.

In the process of rehabilitation, the objectivity and the accuracy of rehabilitation assessment have an obvious impact on the follow-up training. To improve this problem, using a multi-sensor source, this paper attempts to establish a comprehensive assessment method of the finger rehabilitation effect, including three indicators of finger muscle strength, muscle fatigue degree, and range of motion. Firstly, on the basis of the fingertip pressure sensor of the End-Effector Finger Rehabilitation Robot, a mathematical model of finger muscle strength estimation was established, and the estimated muscle strength was scored using the entropy weight method. Secondly, using an sEMG signal sensor, a fatigue monitoring system was designed in the training process, and the fatigue degree was determined on the basis of the change trend of the eigenvalues of MAV and RMS. Lastly, a human–machine motion coupling model was established, and the joint range of motion acquisition and scoring model were obtained on the basis of the motor encoder. According to the above three indicators, using the AHP assessment method to establish a comprehensive rehabilitation assessment method, the effectiveness of the method was verified by experiments. This paper provides a potential new idea and method for objective, accurate, and convenient assessment of finger function rehabilitation, which is of positive significance for alleviating the burden on rehabilitation doctors and improving rehabilitation efficiency.