Aiming at the complex optimization task of collaborative scheduling of multiple types of ground service vehicles in the ramp environment, this paper proposes a deep reinforcement learning algorithm integrated with the Transformer architecture. First, the entire ground service task is decomposed based on the varying priorities of the flight ground service process, transforming the complex multi-type vehicle scheduling problem into a sequential single-type vehicle scheduling problem. The Transformer architecture is then employed to automatically extract the features of flights and vehicles, and task scheduling is solved step by step through the decoder. Preliminary scheduling strategies are generated by combining greedy and Monte Carlo simulation algorithms, which are applied to each sub-problem. On this basis, a deep reinforcement learning algorithm is used to train the entire model, continuously optimizing the scheduling strategy through the interaction between the agent and the environment. Further, to enhance the model’s robustness and ability to handle complex situations, the model is trained by expanding the real dataset. Extensive experiments demonstrate that the deep reinforcement learning approach based on the Transformer architecture effectively prevents mutual interference among different vehicle types and can meet the flight scheduling requirements in real-world environments.