BicNet / CommNet / MARL communication

Hi folks!

Does someone know if there is an BicNet / CommNet or another multi agent algorithm implementation in RLlib where agents are direct communicate with each other and don’t have a central shared Critic, so that the number of agents is scalable?

1 Like

that would be great indeed.

Hi @abb34543, right now I’m implementing one by my self. Watch issue: MARL Custom RNN Model Batch Shape (batch, seq, feature)