Example Policy is very poor about marllib HOT 2 CLOSED

I also ran the configuration provided in another thread

model = marl.build_model(env, mappo, {"core_arch": "mlp", "encode_layer": "128-256"})

This is also very bad with a reward that plateaus very early and does not improve after the first couple of steps.

Is there a setting I can use on any env that will show good results?

from marllib.

Hi,

We have made some revisions to the API usage script. The poor performance you experienced was primarily caused by the absence of force_coop=True. Please give it another try, and you should see that the final reward converges at around 60-70.

By the way, it's important to note that only tasks initialized with force_coop=True are guaranteed to be under a cooperative setting. This is likely the main reason for the poor performance of the previous script, as the agents were not trained to learn teamwork in that setting.

from marllib.