Wumpus-World is a 2D game environment made to test AI agent functionality. The initial world is written by Prof. Simon Parsons (https://scholar.google.co.uk/citations?user=L9z6PakAAAAJ&hl=en), Q-learning functionality + minor bug fixes added 2021.
- Minimap is generated and printed to console (reward map can be as well) IN CORRECT ROTATION!
- Functionality to check whether room is occupied.
- Colored print to console.
- Keep track of gold collected.
- Links expected route functionality (still taking non-deterministic actions into account so might not be accurate) and action to be taken next (UP,DOWN,LEFT,RIGHT).
- What I call 'suicide functionality'. If Link's next goal is closed off by pits or Wumpus', he realizes there is no way out and starts performing random actions.
Run the game using:
python game.py
If on Linux, run like so to produce log file:
python batch.py | tee log.txt
dungeon.py -- draws the dungeon on the screen.
game.py -- runs the game until Link wins or loses.
graphics.py -- simple Python graphics.
utils.py -- utilities used in a few places.
world.py -- keeps track of everything (used by Dungeon to draw).
batch.py -- runs simulations.
Mini-map and Reward-map can be printed to console. Both maps are aligned so that the rewards make sense. Expect a large output.
Grid Size: 6x6-19x19
Enemy Count: 1-2
Pit Count: 1-19
Gold Count: 1-4
Environments: STATIC/DYNAMIC
Outcome: WON/LOST
Death Reason: PIT/WUMPUS
| ENV | OUTCOME | REASON | |
|---|---|---|---|
| Dynamic | Lost | Pit | 10% |
| Dynamic | Lost | Wumpus | 66% |
| Dynamic | Won | - | 22% |
| Static | Lost | Pit | 20% |
| Static | Lost | Wumpus | 2% |
| Static | Won | - | 78% |