Maze Craze

../../_images/maze_craze.gif

This environment is part of the Atari environments. Please read that page first for general information.

Import

from pettingzoo.atari import maze_craze_v3

Actions

Discrete

Parallel API

Yes

Manual Control

No

Agents

agents= ['first_0', 'second_0']

Agents

2

Action Shape

(1,)

Action Values

[0,17]

Observation Shape

(250, 160, 3)

Observation Values

(0,255)

A competitive game of memory and planning!Its a race to leave the maze. There are 3 main versions of the game.

  1. Race: A basic version of the game. First to leave the maze wins

  2. Robbers: There are 2 robbers randomly traversing the maze. If you are captured by the robbers, you lose the game, and receive -1 reward, and will be done. The player that has not been captured will not receive any reward, but they can still exit the maze and win, scoring +1 reward.

  3. Capture: Each player have to capture all 3 robbers before you are able to exit the maze. Additionally, you can confuse your opponent (and yourself, if you are not careful!) by creating a block that looks identical to a wall in the maze, but all players can pass through it. You can only create one wall at a time, when you create a new one, the old one disappears.

The first player to leave the maze scores +1, the other player scores -1 (unless that other player has already been captured in Robbers mode).

Official Maze craze manual. Note that the table of modes has some inaccuracies. In particular, game mode 12 has Blockade enabled, not mode 11.

Environment parameters

Some environment parameters are common to all Atari environments and are described in the base multi-agent environment documentation.

Parameters specific to Maze Craze are

maze_craze.env(game_version="robbers", visibilty_level=0)
  • game_version: Possibilities are “robbers”, “race”, “capture”, corresponding to the 3 game versions described above

  • visibilty_level: A number from 0-3. Set to 0 for 100% visible map, and 3 for 0% visibility map.

Action Space

In any given turn, an agent can choose from one of 18 actions.

Value

Meaning

Value

Meaning

Value

Meaning

0

NOOP

1

FIRE

2

UP

3

RIGHT

4

LEFT

5

DOWN

6

UPRIGHT

7

UPLEFT

8

DOWNRIGHT

9

DOWNLEFT

10

UPFIRE

11

RIGHTFIRE

12

LEFTFIRE

13

DOWNFIRE

14

UPRIGHTFIRE

15

UPLEFTFIRE

16

DOWNRIGHTFIRE

17

DOWNLEFTFIRE

See environment specification to see more information on the action meaning.

Version History

  • v3: Minimal Action Space (1.18.0)

  • v2: Breaking changes to entire API (1.4.0)

  • v1: Fixes to how all environments handle premature death (1.3.0)

  • v0: Initial versions release (1.0.0)