Abstract: Aiming at the impact protection problem of heavy-loaded unmanned platform landing, a buffer airbag system is designed to achieve safe landing. Based on the theoretical model, the airbag geometric parameters and vent area were initially determined, and a finite element simulation model was established using the controlled volume method (CVM) to simulate the airbag cushioning process and analyze the impact dynamics and cushioning performance during the landing attenuation process. Through single parameter analysis, the influence laws of vent hole size, critical exhaust pressure, airbag bottom area and height on cushioning performance are revealed. It is found that there is a conflict between the above parameters reducing the maximum overload and increasing the specific energy absorption. In order to solve this problem, the optimal Latin hypercube design sampling is used, and the neural network is combined to build an objective function proxy model. The accuracy of the constructed proxy model is analyzed, and the NSGA-II genetic algorithm is integrated for multi-objective optimization. The results show that the root-mean-square error between the maximum overload value and specific energy absorption of the proxy model are 0.4895 and 0.7262, and the coefficient of determination (R2) are 0.9833 and 0.9364, both greater than the industry benchmark of 0.9. When the size of the exhaust hole is 3952mm2, the critical exhaust pressure is 158kPa, and the airbag bottom area is 1.08m2, the maximum overload is reduced from 16.8g to 14.5g, and the specific absorption energy is stable at 1529J/kg. Experimental verification shows that the maximum overload test value is 15.2g, and the error from the simulation results is only 4.8%, confirming the reliability of the optimization scheme. This research provides efficient and low-cost design technical support for unmanned platform soft landing systems and improves the safety and efficiency of combat deployment.