Hands-On Reinforcement Learning for Games

Building a Unity environment

The ML-Agents toolkit provides not only a DRL training framework but also a mechanism to quickly and easily set up AI agents within a Unity game. Those agents can then be externally controlled through a Gym interface—yes, that same interface we used to train most of our previous agent/algorithms. One of the truly great things about this platform is that Unity provides several new demo environments that we can explore. Later, we will look at how to build our own environments for training agents.

The exercises in this section are meant to summarize the setup steps required to build an executable environment to train with Python. They are intended for newcomers to Unity who don't want to learn all about Unity to ...