Precision prediction of machined surface roughness is challenging facing the robotic belt grinding of complex blade, since this process is accompanied by significant elastic deformation. The resulting poor prediction accuracy, to a great extent, is attributed to the existing prediction model which less considers the dynamics. In this paper, an improved scallop height model is developed to predict and assess the machined surface roughness by taking into account the elastic deformation and the varying curvature of blade, then robotic belt grinding experiments are carried out to evaluate the proposed model from the perspective of surface roughness. Finally factors that influence the scallop height are analyzed, and the suitable empirical equation of surface roughness is proposed to assess and predict the surface quality from the aspect of blade concave and convex surface by adopting the constant scallop height machining.