Hunt the Wumpus (Logic Agent)

Wumpus World

Create a rational agent program to explore the wumpus world, collecting the gold when possible, but not falling into pits or being eaten by the wumpus. The agent program should use a suitable combination of search and logic techniques.

Performance Measure
The agent program will be evaluated based on the average score over a large number of worlds.

All agent programs will be evaluated in identical worlds.

Environment
The agent is an explorer in a series of connected underground caverns. The agent is equipped with a torch that illuminates only the cavern that is currently being visited. The agent also has a bow and a single arrow. The agent is capable of picking up and carrying the gold treasure.

The world consists of a 4 by 4 grid of connected caverns. Each cavern may contain a pit. If the agent walks into a cavern with a pit, it will fall into the pit and die. One of the caverns will contain a wumpus. If the agent walks into the cavern with the wumpus, the wumpus will eat the agent.

If the wumpus is dead, the agent will not be eaten by entering the cavern containing the wumpus.

• The world is partially observable. The agent must explore to discover the location of pits, the wumpus and the gold.
• The world is static. Objects only move if the agent acts upon them.
• The exploration of the world is sequential. The agent may learn more about the world as it explores. Each new world exploration is episodic in nature.
• The environment is deterministic.
• The environment is discrete in time and space.
• The environment is a single-agent environment.

Actuators
The agent has the following actions:

• TURN_RIGHT
• TURN_LEFT
• FORWARD The agent is always facing in one of 4 directions. A TURN_RIGHT will change that direction 90 degrees clockwise. A TURN_LEFT will change that direction 90 degrees counter-clockwise. A FORWARD will move the agent in the direction it is facing, unless a wall is in the way.
• GRAB The agent will pickup the gold if it is in the same cavern.
• SHOOT The agent will shoot its single arrow (if it has not already been shot) in the direction it is facing. The arrow will travel from the current cavern, through all caverns in the facing direction until it hits the wumpus or a wall. The wumpus will die if it is hit by the arrow.
• QUIT The simulation will be terminated. The agent should execute this action when the gold has been picked up, or when the agent has decided that further exploration is too dangerous.
• NOOP The agent should never need this operation.

Sensors
The agent receives the following percepts

• ARROWS 0|1 The number of arrows the agent is holding.
• GOLD 0|1 The number of golds the agent is holding.
• BUMP T Only received if the agent bumped into a wall.
• GLITTER T Only received if the gold is in the same cavern as the agent.
• BREEZE T Only received if one or more of the neighboring caverns contains a pit.
• STENCH T Only received if one or more of the neighboring caverns (or the current cavern) contains the wumpus.
• SCREAM T Only received if the wumpus was killed as a result of shooting the arror in the previous action.

Suggested points of interest

• The DPLL_Entails method should be used to improve performance of the inference engine.