Collaborative Game Challenge

[bahriddin]

Minecraft Mafia

There are 2 teams in M x N world. Initially, team agent's position, position of
walls are placed symmetrically. Each team has same K number of agents. Each
team has to kill enemies. Killing condition is following:

• If both team player's are looking face-to-face, one of them randomly will
  be dead.
  
• Otherwise, who facing towards to the enemy and adjacent to it will kill.
  

Some of the complex situations:

All of these actions have the same price:

• move forward, backward, right, left
• turn left
• turn right

So turning back takes 2 steps: whether 2 times left or 2 times right.
Each team's goal to remove all members of competitors. Each team members will play turn-by-turn. So if we call teams A and B: A1-B1-A2-B2-...
If none of the teams won't win after (N+M) x K x c (c=const) moves, the game will be stopped and draw result would be recorded.

This is partially observable game: agents have exact knowledge about what they see and if enemy is closer than 4 blocks, they hear and guess enemy is close.

Questions

1. Can AI play better than humans?
2. Which algorithms show best results to solve this problem: to learn faster, to show better results, more robust.

Hypothesis

In simple shooting/fighting complex world team games (like Counter Strike, Quake or tank), can we create smart AI team to beat humans.

[bahriddin]

TBD:

• Create transition function
• try to show computational complexity