Question Strikers Vs. Goalie learning

Hi, Goalie and Strikers have the same scripts, but different neural networks and behavior names.
How were they trained?
In the documentation, I found only "self play", but for commands that perform the same actions. In this case, the commands have different actions. And such a "self play" leads to one neural network.

I tried to run two consoles to train two different neural networks, but one of them failed to connect. Do I need to use different ports?

https://github.com/Unity-Technologies/ml-agents/tree/main/Project/Assets/ML-Agents/Examples/Soccer

 

Have you looked into the config? There should be two different agents listed. Start training with one console.
to be honest I dont know if that produces two brains or just one

 

Thanks, found the config and everything became clearer

config:
https://github.com/Unity-Technologies/ml-agents/blob/main/config/poca/StrikersVsGoalie.yaml

docs:
https://github.com/Unity-Technologi...erative-multi-agent-environments-with-ma-poca
https://github.com/Unity-Technologi...gn-Agents.md#groups-for-cooperative-scenarios

 

glad I could help you