In recent years, deep reinforcement learning has garnered significant attention because it can be applied to higher-dimensional environments compared with traditional reinforcement learning. However, the number of trials increases in behavior acquisition, particularly in tasks with high dimensions and sparse rewards. To improve learning speed, we apply curriculum learning, which improves the learning performance by changing the difficulty of the task in a stepwise manner, to the behavior acquisition of a shooting game as well as conduct experiments. We compare the learning performance with and without the application of curriculum learning and confirm the faster behavior acquisition of the shooting game AI through experimental evaluation. Additionally, we analyze and discuss the development of other tasks and an algorithm for automatic curriculum generation.