Of all the chapters in this book, this may be the most useful if you are in the process of developing your own game. Game testing is one of those things that requires so much time and attention, it has to be up for some form of automation. While it makes sense for DRL to work well in this area for almost any game, it remains to be seen whether that is one of the niches for this new learning phenomena. One thing that's for sure, however, is that ML-Agents is more than capable of working as a testing harness, and we are sure that it will only get better over time.

In this chapter, we looked at building a generic testing platform, powered by ML-Agents, that we can use to test any game automatically. We first looked at each of the components that we needed to adapt, the academy and the agent, and how they could be generalized for testing. Then, we looked at how we could inject into the Unity input system and use our TestingAgent to override the game's input and learn how to control it on its own. After that, we looked at how to better set up our testing by using offline IL and recording a demo file that we could use to train the agent later. Finally, in order to see how well our testing was doing, we added analytics and customized them to our needs.

The next chapter will be our final chapter and our last discussion of deep learning for games; appropriately enough, we will look at what the future holds for ML-Agents and DRL.