Adversarial and cooperative self-play

The term self-play can, of course, mean many things to many people, but in this case, we mean the brain is competing (adversarial) or cooperating with itself by manipulating multiple agents. In the case of ML-Agents, this may mean having a single brain manipulating multiple agents in the same environment. There is an excellent example of this in ML-Agents, so open up Unity and follow the next exercise to get this scene ready for multi-agent training:

1. Open the SoccerTwos scene from the Assets | ML-Agents | Examples | Soccer | Scenes folder. The scene is set to run, by default, in player mode, but we need to convert it back to learning mode.
2. Select and disable all the SoccerFieldTwos(1) to SoccerFieldTwos(7) areas. We won't use those yet.
3. Select and expand the remaining active SoccerFieldTwos object. This will reveal the play area with...