Is it possible to train more than one policy in the same Emulated environment?

[camillo2008]

I am sorry if my question is confused or completely nonsense. I am a little bit familiar with SB3, I am wondering if extending BaseAviary(or some base class) class it is possible to create one environment where two group of drone compete in some way. For example let's say we have the "predator drone" and the "pray drone", those two group need to have different policy, but in the same time they need to be placed in the same shared environment. I am wondering, is it possible to use VecEnv, with different drone policy, like in the example before? May you share some hint if this is possible or this isn't the case, and I am completely missing the point.

Thank you

[JacopoPan]

It's technically doable but it requires to develop at least some additional lightweight interface/wrapper. It also depends on whether

• you want to have 2 (adversarial) drones
• you want to have 2 (adversarial) groups of arbitrary sizes
• whether both or only one of the teams is a learning agent (and in the first case, how you interleave the learning or, in the second, what is the policy for the non-learning team)

[camillo2008]

I was thinking of two teams(of arbitrary size) learning at the same time, however thanks for taking the time to answer me, now that I know that is possible to train two policy in the same environment I will try to carry on the idea. If you have any other hint that you want to share fell free to write it, otherwise you can close this issue, and thank you so much