Abstract: A multi-function rescue attachment with tonging, shearing, and grabbing capabilities was developed to improve the rescue operation efficiency and save the time of switching different attachments during rescue operations. The weighted least squares support vector machine (WLS-SVM) response surface method was improved to establish the approximate model of grabbing mechanism. The maximum equivalent stress, the maximum deformation and the mass were proposed to evaluate the performance of mechanism based on the approximate model, then the non-dominated sorting genetic algorithm (NSGA-Ⅱ) was introduced for multi-objective optimization. To improve the population diversity and the search ability of NSGA-Ⅱ, the elite strategy, the crossover operator and the mutation operator were improved. Additionally, the NSGA-Ⅱ and the improved algorithm were used for multi-objective optimization of grabbing mechanism. Finally, the multi-dimensional Pareto front was visualized in a two-dimensional plane by the radial coordinate visualization method, and the most satisfied solutions were selected. The comparison shows that the Pareto solution distribution of the improved algorithm can be distributed more uniformly, and the target values of the most satisfied solution can be smaller. Particularly, the lightweight design can be effectively carried on in a condition that ensures the reliability of the grabbing mechanism.