Gridworld environments for Open AI gym

Maps

• Puddle World ('PuddleWorldB-v0','PuddleWorldST1-v0', 'PuddleWorldST2-v0',etc)
• Mine World ('MineWorldRandomSmall-v0','MineWorldRandomBig-v0')
• Room World ('RoomWorldFinalBig-v0', 'RoomWorldObjectSmall-v0','RoomWorldExit-v0',etc)

and (not changed from starting repo)

• Four Rooms
• Simple Gridworld

Example Maps

To install

cd gym-grid
pip install -e .

To use

import gym
import gridworlds

env = gym.make('RoomWorld-v0')

To test

cd tests/
test_env.py

Recommended to run tests in an python terminal.

Map Descriptions

• PuddleWorld - Avoid centre of map and reach goal (goal position varies depending on map being B,ST1,ST2,etc)
• MineWorld - Maximise reward by avoiding Mines and collecting Fruits. Map is randomised at every call of env.reset()
• RoomWorld - Complete episode by collecting fruit (to train an 'collect fruit' policy )
• RoomWorldExit - Complete episode by reaching gap from random start (to train an 'exit room' policy )
• RoomWorldObjectSmall - Complete episode by collecting the fruit and reaching gap from random start (to train an 'collect fruit and exit room' policy )
• RoomWorldFinalBig - Complex Map with 6 rooms. Collect all fruits to finish. (Meant as a hierarchical RL benchmark)

Note

• env_wrapper.py is crucial to get square view output around the agent
• all the env ids are in gridworlds/__init__.py
• Roomworlds have the centre of view being number of fruits in a particular room

Starting Point being this repo