Best Practices for Training a Team with Different Roles

Hello, let's use the Soccer project as an example. If I want to train one of the agents to behave like a striker (getting a reward for each goal) and the other agent to behave like a goalkeeper (staying close to the goal and getting a reward for each save). Can I train them together using some parameters or variables? Or is the best way to achieve this goal to train one of the agents first and then train the other agent separately?

 

I have not tried it yet. But as an option, you can program two different agents differently with different brains. And then they will train together but in different ways. Just give a different name to the brain and different reward.