Question Initializing MA-POCA such that different team members have different policies

Hi,

I have created a game with 2 teams and 3 members in each team. Each team tries to eliminate the other by shooting balls at them (similar to the DodgeBall environment in Unity). I have found MA-POCA to work well for this scenario. However, I want to train my agents with different capabilities, and hence perform differently. How do I set up my .yaml file such that i can train agents using MA-POCA such that each agent in a team can have a different policy.

 

You can configure different Behaviours in your yaml file, but I think one of the limitations of POCA is that you can have only one behavior per team.

Code (CSharp):

1. behaviors:
2.   Agent1:
3.     trainer_type: poca
4.     hyperparameters
5.     ...
6.   Agent2:
7.     trainer_type: poca
8.     ...
9.  
10.