AI multi-agent search

Introduction

Pacman finds himself in a world filled with ghosts relentlessly pursuing him. His objective is to evade his enemies by consuming all the available food and reaching the final destination. To achieve this, he requires a mechanism to devise a strategy. In this project, I have implemented the reflex agent, minimax agent, and expectimax agent algorithms for Pacman to employ and survive in this world.

Final result

Reflex agent

A reflex agent operates based on simple condition-action rules or “reflexes.” It takes the current percept, which represents the agent’s current state of the environment, and directly maps it to an action without considering the history or future consequences. The following demonstrates Pacman acting as such an agent.

Minimax agent

This agent determines the optimal move by evaluating all potential outcomes through recursive evaluation of the game tree. The following demonstrates Pacman utilizing this algorithm with α-β pruning.

Expectimax agent

The minimax agent always acts optimally, assuming opponents consistently act aggressively in an attempt to eliminate their foes. In contrast, an expectimax agent considers the possibility that adversaries may make mistakes or suboptimal moves. The following demonstrates the expectimax agent, which takes a more cavalier approach by assuming adversaries do not always choose their best actions.

_{Reference to the stub code}