Is this actually Partial Brute Force? about multi-agent-emergence-environments

[acyclics]

From what I understand about Attention, it's a way to encode inputs into a format that's "relevant" to the neural network. I guess Attention helps by learning a format for entities (boxes etc) that helps the agent decide what's the best action to take (i.e. where to move the box). Without first learning a relevant format for these entities, perhaps said "brute-force" approach would not be sufficient for the agent to find these best actions randomly.

[bladedsupernova]

Still lost.

How does hider1 move the box in place for when hider2 arrives back and hider1 then goes to get his box. This was learnt when OpenAI purposely moves the ramp to the back of the room.

I realize they learn general tricks but this seems like brute force too.