search

google-deepmind / lab

A customisable 3D platform for agent-based AI research
Other
7.06k stars 1.36k forks source link

Make random maze as Capture The Flag mode #225

Open kimbring2 opened 2 years ago

kimbring2 commented 2 years ago

Hello,

Currently, I am trying to implement paper 'Human-level performance in first-person multiplayer games with population-based deep reinforcement learning'.

I can make a simple 1 vs 1 CFG game using the GtkRadiant and Lab.

However, it seems like paper use a Maze generation method to make various environment. It is little hard to make it myself because of lack of information.

Would you provide little hint about how to build a CFG game using Maze generation if it is possible?

Thank you

tkoeppe commented 2 years ago

Hm, yes, we used an internal tool to generate mazes I think. It's pretty similar to what you can do with text-based levels, though, so maybe you can just emit random text levels and then compile those?

@charlesbeattie: any more ideas?

kimbring2 commented 2 years ago

@tkoeppe Thank you for response. I can make a random maze from text.

I am wondering how can I put a component for CFG game such as the info_player_intermission, team_ctf_blueflag, team_ctf_redflag, team_ctf_blueplayer, team_ctf_redplayer, team_ctf_bluespawn and team_ctf_redspawn into text map as character symbol.

I could select them from GUI menu at GtkRadiant.