top of page

PROJECT AND ROUND-UP OVERVIEW

The Contest, The Robots, and How They Function

The Wumpus World Problem:

On a rectangular grid, there exists one Wumpus, one pile of gold, a non-negative number of pits, and one Agent that starts at the bottom-left of the grid. The goal of the Agent is to reach the space that contains the pile of gold and return to its origin, all without falling into a pit or occupying the same space as the Wumpus. The Agent uses its 'senses' to determine the objects around it. A stench, breeze, or glitter correspond to a Wumpus, a pit, or a pile of gold next to the Agent, respectively. The Agent must use deductive reasoning to reach the pile of gold; grids are designed so that the Agent never needs to guess where to go next. Also, the Agent has one arrow to shoot and kill the Wumpus with. In some cases, killing the Wumpus is required to reach the pile of gold!


The Agent—ARA-24
The Agent first analyzes its senses and memorizes all the possible objects (e.g. the Wumpus, a pit, or the pile of gold) that could occupy an adjacent space. For example, if the Agent doesn't sense anything, it can deduce that all of the adjacent spaces are empty. In some cases, the Agent receives contradictory information. For example, the Agent may sense a pit at a particular space, then it may sense the pile of gold at the same space. Therefore, the Agent can conclude that this space must be empty since it cannot contain both a pit and the pile of gold.

​

Once the Agent reasons about all the possible objects on its adjacent spaces, it must determine its next action. If the Agent is in an adjacent space to the pile of gold, it will immediately move to that space. It will then attempt to find the shortest path back to its origin, without having to do any further exploration and reasoning. If the Agent is in an adjacent space to the Wumpus, it will shoot it. If it is not, the Agent will look for all the safe spaces that are adjacent to it. If there's only one safe adjacent space, the Agent will move to it. If there are multiple safe spaces, the Agent will choose a space that it hasn't visited previously. If all the safe spaces have been visited, it will choose the space it has visited the least recently.

​

The Round-Up:
During the round-up, several Agents attempted to solve five different grids, each being more difficult than the last. On each grid, there were a couple of Agents that ran into a hazardous object (e.g. a pit or the Wumpus), got stuck in an infinite loop, or crashed in the middle of their algorithm. Our Agent was one out of the three Agents that successfully found the pile of gold in all five grids.

However, our Agent did not have the most effective algorithm for finding the pile of gold and backtracking. Hence, our Agent took more steps to complete the grid compared to the others. One of the Agents surprised many participants (including Professor Bram!); while backtracking to its origin, the Agent did more exploring and found a shortcut to the origin. Our Agent did not take this shortcut, as it didn't deem the spaces in the shortcut as safe during its exploration. The fact that their Agent explored after obtaining the pile of gold led it to find a more efficient path. Some Agents were programmed to recursively find the shortest path to the origin once it obtained the pile of gold. This, albeit taking more computing time, led to a more efficient path back to the origin.

​

Is ARA-24 Intelligent?
While the performance of ARA-24 may seem like magic to those not as versed in computer programming and logic, it is nowhere near the level of computing that is considered Artificial Intelligence. ARA-24 cannot think or reason about anything other than exactly what we, the programmers, present to it. Even though it seems to perfectly avoid the hazardous objects scattered around a grid, it does not comprehend why it avoids these objects. ARA-24 does not understand that if it occupies a space with a pit or the Wumpus, it will fail. The only reason why it seems to actively search for the pile of gold is because we programmed it to move to it. ARA-24 is programmed to shoot the Wumpus if they are adjacent to each other. A sentient being would have some reservations on the morality of killing the Wumpus in cold blood. ARA-24, however, has no sense of morality, and only does what it deems to be "efficient" and "logical."

Overview: Intro
bottom of page