Abstract: To overcome the influence of the interaction forces of human-robot on the trajectory tracking performance of rehabilitative training walker, a novel design method of observer was proposed based on position output of system to estimate the interaction forces of human-robot. Firstly, a nonlinear controller was designed and a system model was established for every motion axis to restrain the interaction forces and avoid larger tracking errors, in order to improve the system's tracking performance and guarantee the rehabilitee's safety. Further, taking the predictive control as a method to constrain the trajectory and velocity tracking errors simultaneously, the optimal trajectory tracking was realized for every motion axis. Finally, a simulation and experiment were carried out to analyze the running effect of the walker. Results show the effectiveness and advantage of the observation of human-robot interaction forces and the design method of controller.