In MARL (Multi Agent Reinforcement Learning), is it possible for agents to move simultaneously and take different actions?

[G-JOEY]

Probably the exact same question as #906 .

I'm trying MARL in an environment where agents take actions simultaneously, but may do different actions, like in MOBA games.

Looking at (https://tianshou.org/en/stable/01_tutorials/07_cheatsheet.html#multi-agent-reinforcement-learning), it says "Simultaneous move: at each timestep, all the agents take their actions (example: MOBA games)" and "For simultaneous move, the solution is simple: we can just add a num_agent dimension to state, action, and reward. Nothing else is going to change.", but I'm having a hard time finding examples or information on that.
(Of course, this could be due to my lack of understanding of Tianshou. Or it could be that the description implies that the agents move simultaneously but take the same action, although I don't think so.)

I have verified that Tianshou works perfectly in a Turn-based environment like Tic-Tac-Toe, but as I said above, I need MARL in an environment where the agents move simultaneously and can take different actions.

Maybe it's because all the examples I can refer to are Turn-based environments, but even in my custom environments, the agents always take the same action.

Does Tianshou provide any way to allow each agent to have their own action rather than a shared action for the Simultaneous Move case?

If it's not too much trouble, please add a simple example or reference document along with your answer.

Thank you.