Quadruped robots, with their multi-legged structure, are highly mobile and capable of navigating in uneven terrains and environments with obstacles. Recently, Transformers have gained significant attention as large-scale language models. Due to their ability to efficiently process time-series data, Transformers are also useful in robotic action planning. By combining Transformer-based action planning with the mobility of quadruped robots, it is believed that they can be applied in various scenarios, such as construction sites and disaster-stricken areas. This paper proposes a behavior control method for quadruped robots based on task instructions using Transformers and deep reinforcement learning. The proposed quadruped robot behavior control model was tested in a simulation environment where multiple objects with different colors and shapes were placed and the task was to approach a specified object. The evaluation results have shown that the overall average task success rate was 80.82 [%].