Abstract: A method that enables rehabilitation robots to train affected limbs by referencing the motion postures of healthy limbs is proposed in this study. This method ensures that the trajectory of the rehabilitation robot aligns very closely with the target patient′s motion habits, personalizes rehabilitation training, exercises the coordination of the patient′s bilateral limbs, and improves the effect of the training. In this study, a four-degree-of-freedom pneumatic upper limb exoskeleton rehabilitation robot is designed, and a master-slave heterogeneous motion-mapping algorithm is proposed to deal with the different degrees of freedom of the robot and upper limb. The exponential product method is used to establish the robot kinematic model, define the equivalent physiological joints of the robot, and perform position mapping between the healthy limb and the robot. For joint mobility training, a position mapping algorithm characterized by quaternion-based pose and mirror solving is developed to realize equivalent human-robot joint mapping by leveraging the independent kinematic joint poses of healthy limbs to improve the mapping efficiency. In the activities of daily living training involving multiple joints, the motion mapping method for elbow-wrist separation is proposed. The proposed motion-mapping method solves the rotational motions of two joints at the shoulder and one joint at the elbow of the robot using the vectorial motions of the large and small arm rigid bodies of the healthy limb, respectively, and ensures positional accuracy during motion mapping. Finally, rehabilitation motion following and position similarity verification experiments are conducted to verify the feasibility of the master-slave heterogeneous motion mapping method.