The fluency and intelligence of human-computer interaction affect the working efficiency of robots and the operator"s operating experience. In order to improve the fluency of human-computer interaction and the understanding ability of operator"s command statements, a design method of human-computer interaction system based on large language model was proposed. This method mainly studies the combination of large language model and robot automatic control. To solve the problem that the traditional speech recognition module can recognize only a limited number of command statements and has strict requirements on the complexity of command syntax and sentence length, the large language model is introduced as the core of command understanding. In order to solve the problem of complicated and rigid interaction process in traditional human-computer interaction process, the human-computer interaction process of large language model and decoder to generate subtask sequence is proposed. Simulation tests show that this method can effectively improve the performance of man-machine collaboration tasks, and the pre-trained large language model can understand operator commands in detail and give corresponding feedback, which is helpful to build a more natural and intuitive human-machine interaction.