Aiming at the problems of insufficient leg stiffness and vibration of large hexapod robots during motion， a multi-objective comprehensive optimization method based on an approximate model is proposed. First， to determine the optimal space for the bionic leg， a finite element model is established to analyze the strength， stiffness， and modal frequency of the leg structure under various complex working conditions. A parametric model is established for the static and dynamic performance of the bionic leg， and the corresponding design variables are defined. To obtain the initial sample points， the Optimal Latin Hypercube Method was used to conduct an experimental design of the bionic leg hip linkage， thigh， and calf module. The response surface model， kriging model and radial basis function neural network model are fitted， and the approximate model with the highest accuracy is selected through error analysis. And combined with the Non-dominated Sorting Genetic Algorithm Ⅱ， the static stiffness， mass， and first-order natural frequency are targeted.Constraining the maximum stress， the bionic leg is optimally designed， and the optimized results are analyzed and verified. The results show that after the heavy-duty bionic leg is optimized by the method， under the premise of satisfying the structural strength， the maximum deformation of the flat ground condition is reduced by 9.73%， the maximum deformation of the slope condition is reduced by 9.46%， the first-order natural frequency is increased by 3.45%， and the overall quality fell by 8.63%.