Industrial robot-assisted abrasive cloth wheel (ACW) accurately polish blades is considered to be a challenging task, and it is necessary to realize the digitalization of the process. Due to the flexible contact characteristics of abrasive cloth wheel and the change of blade surface curvature, the amount of microscopic material removal and the topographical errors of the blade surface are not uniform. So the surface roughness value is larger. In this paper, considering the flexible deformation of the blade and the abrasive cloth wheel during polishing contact, the polishing contact model of the blade and the abrasive cloth wheel is established and simulated. Firstly, based on the Preston equation and Hertz contact theory, the material removal profile in the contact area of the blade and the abrasive cloth wheel is analyzed. Secondly, the step length is optimized by considering the deformation of the abrasive cloth wheel in thedirection, and the line spacing is optimized by considering the material removal uniformity in thedirection. The NURBS curve is used to extract the blade polishing area curve and generate the polishing trajectory data points. Then, two methods of rigid trajectory planning and flexible adaptive trajectory planning of abrasive cloth wheel are simulated and analyzed by offline programming software. Finally, experiments were carried out on a four-station wheel changing polishing platform. Simulation and experiments results show that the proposed flexible adaptive trajectory planning method can make the surface roughness of the convex and the concave, the surface roughness of the leading and trailing edge, the total polishing efficiency increased by about 9.4%.