This project takes snooker as the research object and tries to explore the whole process of applying machine learning to the Unity project through reading literature and combining it with experiments. Because snooker game combines the characteristics of physical movement, turn-based system, strategy confrontation, and so on, the analysis of it can cover considerable types of game categories, so it has a certain practical value. As the result, this report summarizes some prominent problems and matters needing attention in the process of project development according to experimental data. Including but not limited to the discussion of the training cost, reward mechanism, convergence, and unbiased environment building. These contents should be able to provide a reference for the developers who are interested in using Unity Engine and the ML-Agents package but lack relevant experience. Also, the report concludes with an idea for improving the SAC policy. It may be useful for algorithmic researchers.