Wumpus World

Simple simulation based on the book Artificial Intelligence: A Modern Approach, by Russell and Norvig.

Open developer console to check logs.

Seed: ?
Score: ?
?
?
?
Mouse controls
  • Use this panel to control the game and AI.
  • Click game area to focus (for keyboard).
  • Left/Right click game area to mark tiles (visual only).
Keyboard controls

Requires game focus:

  • WASD/Arrows to move.
  • SPACE to shoot arrow.
  • N to create new world.
  • R to reset current world.
  • G to move using the agent.
  • Other actions are automated.
Score rules
  • Any of the actions FORWARD, TURNLEFT, TURNRIGHT, GRAB: -1 pts
  • SHOOT arrow: -10 pts
  • CLIMB without gold: -100 pts
  • CLIMB with gold: +1000 pts
  • Fall or get eaten: -1000 pts
  • The best score is 995 points.
AI idea

The AI agent's goal is to retrieve the gold maximizing the score.

At first, the agent knows about the basic rules (e.g. pits are surrounded by a breeze), that the start square is safe, that it is facing east, that there is one gold to get, and the wumpus to avoid.

The agent uses these percepts to learn about the environment: BREEZE, STENCH, GLITTER, BUMP, SCREAM.

Last batch results
  • Run count: ?
  • Elapsed time (s): ?
  • Score average: ?
  • Score median: ?
  • Gold percent: ?