WFC¶


Action Space	`Discrete(7)`
Observation Space	`Dict('direction': Discrete(4), 'image': Box(0, 255, (7, 7, 3), uint8), 'mission': MissionSpace(<function WFCEnv._gen_mission at 0x7f756a3278b0>, None))`
Creation	`gymnasium.make("MiniGrid-WFC-MazeSimple-v0")`

Description¶

This environment procedurally generates a level using the Wave Function Collapse algorithm. The environment supports a variety of different level structures but the default is a simple maze. See WFC module page for sample images of the available presets.

Requires the optional dependencies imageio and networkx to be installed with pip install minigrid[wfc].

Mission Space¶

“traverse the maze to get to the goal”

Action Space¶

Num	Name	Action
0	left	Turn left
1	right	Turn right
2	forward	Move forward
3	pickup	Unused
4	drop	Unused
5	toggle	Unused
6	done	Unused

Observation Encoding¶

Each tile is encoded as a 3 dimensional tuple: (OBJECT_IDX, COLOR_IDX, STATE)
OBJECT_TO_IDX and COLOR_TO_IDX mapping can be found in minigrid/core/constants.py
STATE refers to the door state with 0=open, 1=closed and 2=locked (unused)

Rewards¶

A reward of ‘1 - 0.9 * (step_count / max_steps)’ is given for success, and ‘0’ for failure.

Termination¶

The episode ends if any one of the following conditions is met:

The agent reaches the goal.
Timeout (see max_steps).

Registered Configurations¶

MiniGrid-WFC-MazeSimple-v0
MiniGrid-WFC-DungeonMazeScaled-v0
MiniGrid-WFC-RoomsFabric-v0
MiniGrid-WFC-ObstaclesBlackdots-v0
MiniGrid-WFC-ObstaclesAngular-v0
MiniGrid-WFC-ObstaclesHogs3-v0

Note: There are many more unregistered configuration presets but they may take a long time to generate a consistent environment.

They can be registered with the following snippet:

import gymnasium
from minigrid.envs.wfc.config import register_wfc_presets, WFC_PRESETS_INCONSISTENT, WFC_PRESETS_SLOW

register_wfc_presets(WFC_PRESETS_INCONSISTENT, gymnasium.register)
register_wfc_presets(WFC_PRESETS_SLOW, gymnasium.register)

Research¶

Adapted for Minigrid by the following work.

@inproceedings{garcin2024dred,
  title = {DRED: Zero-Shot Transfer in Reinforcement Learning via Data-Regularised Environment Design},
  author = {Garcin, Samuel and Doran, James and Guo, Shangmin and Lucas, Christopher G and Albrecht, Stefano V},
  booktitle = {Forty-first International Conference on Machine Learning},
  year = {2024},
}