Welcome to the Wumpus World problem, a classic artificial intelligence challenge. In this problem, an agent must navigate a four by four grid to find gold while avoiding deadly hazards. The agent starts at position one one in the bottom left corner. The goal is to find the gold, shown as a star, and return safely to the starting position. However, the world contains dangerous obstacles: a Wumpus monster that will kill the agent if encountered, and bottomless pits that are equally deadly. The challenge is that the agent begins with no knowledge of where these hazards are located.
The agent navigates using five types of sensory information called percepts. First, a stench is detected in rooms adjacent to the Wumpus, shown here as green clouds. Second, a breeze is felt in rooms next to pits, indicated by blue wind symbols. Third, glitter appears when gold is in the current room, shown as sparkles. Fourth, a bump occurs when trying to move into a wall. Finally, a scream is heard when the Wumpus is killed by an arrow. The agent must use logical reasoning to infer safe paths from these limited clues, as it cannot directly see hazards until it encounters them.
The agent has five basic actions available. It can move forward one room at a time, turn left or right to change direction, shoot an arrow to kill the Wumpus, grab gold when in the same room, and climb out from the starting position. The agent follows a systematic exploration strategy, moving only into rooms it believes are safe based on its percepts. Here we see a possible safe path shown in green, where the agent explores room by room, eventually shooting an arrow to eliminate the Wumpus threat, then proceeding to collect the gold and return home safely.
The agent uses logical reasoning to build knowledge about the world. It applies inference rules based on its percepts. For example, if no stench is detected, the agent can safely conclude there is no Wumpus in adjacent rooms. Conversely, detecting a stench means the Wumpus is nearby. The agent maintains a knowledge base using propositional logic, marking rooms as safe shown in green, dangerous shown in red, or unknown shown in yellow. Through systematic logical inference, the agent can deduce which rooms are safe to explore and which should be avoided, allowing it to navigate efficiently toward its goal.
To summarize what we have learned about the Wumpus World problem: It demonstrates how artificial intelligence agents can reason under uncertainty using logical inference. The agent builds a knowledge base from sensory percepts and applies systematic exploration strategies. This classic problem illustrates fundamental AI concepts including planning, knowledge representation, and logical reasoning in uncertain environments.