Escaping the harsh realities of the year 2020 has taken on various forms. For some, the year 2020 has been rife with reading, podcast listening, or late-night binging of the latest Netflix series. For me, nothing says departure from reality better than building an artificially intelligent agent to outperform a duo of ghosts who are hell-bent on eating it. Yes, you guessed correctly, I’m talking Pac-man and I’m talking AI. Let’s get started!
The game will be played on a tiled game board containing both our characters and objects. Time in this context works in a discrete manner rather than continuous, and as a consequence, the following discussion will refer to time in step (e.g. the kth time step). Within this world, there exists:
Figure 1: Pac-man game board (Image by Author).
The interactions between these characters and the world around them are encapsulated within a complete board representation of the game — otherwise known as the state. Each state represents the world at a specific point in time and any action performed by either Pac-man or the ghosts will result in a change of state. The actions available to either Pac-man or the ghosts are movement-based actions that dictate whether the entity moves up, down, left, or right. Furthermore, like any good game, there exists a point system that is altered slightly from the traditional game.
#mathematics #pacman #reinforcement-learning #data-science #inside-ai #ai