[WIP] Pettingzoo Multi-Agent Wrapper #649
Conversation
…-pettingzoo Add PettingZoo parallel wrapper for cooperative multi-agent VizDoom
mwydmuch
left a comment
mwydmuch
left a comment
There was a problem hiding this comment.
Need to tests it more, I'm unsure about respawn mechanic.
But I think this is looks good.
What do you think is missing to call it good first version with this one environment?
There was a problem hiding this comment.
What was the change here? I would prefer not to change the scenario we have for such a long time.
| }) | ||
| steps += frames_per_step | ||
|
|
||
| elif cmd == "respawn": |
There was a problem hiding this comment.
What if there is respawn delay?
| if is_dead: | ||
| if verbose: | ||
| print(f"Player {agent} respawning at step {game.get_episode_time()}...") | ||
| game.respawn_player() |
There was a problem hiding this comment.
Instead of calling game.respawn_player() that will block the current controller, it is also possible to send "use" action and wait (advance_action) or (make_action).
|
I've addressed all the points you've raised. I am not sure about the respawn part. It seems to work fine as is. Needs further testing to see whether there are situations where issues can arise |
3 participants
pettingzoo_wrapper/base_pettingzoo_env.pycontains the multi-agent environment wrapperpettingzoo_wrapper/__init__.pyis the main entry point for creating envsexamples/python/multi_agent.pyruns a few episodes with rendering for quick validationexamples/python/train_multi_agent.pyuses the benchmarl library to train a multi-agent policy