05-22-2016 at 02:28 PM
Part 3 - Ghost Behaviour - Research done on pacman and mspacmanArticle: "Evolving a Ms. Pac-man controller - using grammatical evolution"
"This game, the sequel to Pac-Man, consists of guiding Ms. Pac-Man through a maze, eating pills, power pills, and fruit. This task would be simple enough if it was not for the presence of four ghosts that try to catch Ms. PacMan. Each ghost has their own, well-defined, behaviour. These behaviors are the largest difference between the Pac-Man and Ms. Pac-Man. In the original PacMan, the ghosts are deterministic and players who understand their behavior may always predict where the ghosts will move. In Ms. Pac-Man, the ghosts have non-deterministic elements in their behavior and are not as predictable."
Article: Learning to Play Using Low-Complexity Rule-Based Policies: Illustrations through Ms. Pac-Man
"In the original version of Pac-Man, ghosts move on a complex but deterministic route, so it is possible to learn a deterministic action sequence that does not require any observations. Many such patterns were found by enthusiastic players. In most of Pac-Man's sequels, most notably in Ms. Pac-Man, randomness was added to the movement of the ghosts. This way, there is no single optimal action sequence, observations are necessary for optimal decision making. In other respects, game play was mostly unchanged."
Article: "Discovering Multimodal Behavior in Ms. Pac-Man through Evolution of Modular Neural Networks"
"Even the original Pac-Man is a poor choice for AI research. Ghost behavior is deterministic, so it is possible to maximize the score by following memorized paths, without any strategic intelligence. For this reason, current research focuses on Ms.Pac-Man, which is non-deterministic. Non-determinism makes evaluations noisy, which in turn makes learning hard. Another difference is that Ms. Pac-Man has four mazes in comparison to Pac-Man’s one.
Because of these differences, success in Ms. Pac-Man depends more on generalization than memorization"
Research Paper: Machine Learning Applied to Pac-Man
David Bigas Ortega
Director: Javier B´ejar Alonso
Barcelona School of Informatics - FIB UPC
The following aspects need to take into consideration when analyzing the AI of ghost behaviour in pac-man and ms. pac-man respectively.
Number of characters (the four ghosts and Pac-Man).
• How many number of different mazes
• How many number of stages till the end of the game
• Different items that can be eaten (dots, energisers and fruits).
• Restrictions in the movement of the characters in different context there current movement towards their expected movement vs actual movement
• Difference in speed between Pac-Man / Ms. Pac-man and the ghosts.
• Reduction / Increase of the ghosts’ speed and the Pac-Man’s speed when Pac-Man eats and the Ms. Pac-Man’s speed when Ms. Pac-Man eats and the an energiser.
• Delay time to the ghosts’ departure from the ghost box.
• Different targets tiles for each ghost as the Pac-Man vs Ms. Pac-Man moves or standing still
• Reducing speed when Pac-Man vs Ms. Pac-Man when dots get eaten.
• Reducing speed when the ghosts are in the tunnel or ghosts at the same speed as Pac-Man / MS. Pac-Man
• Behaviour when Pac-Man turns left or right vs when Ms. Pac-Man when in the same position towards the ghosts.
• Original system collision between Pac-Man and the ghosts vs Ms. Pac-Man and the ghosts - Are their more "boeys" in Pac-Man than Ms. Pac-man , why....?
• How far a Pac-man/Ms.Pac-man from the ghosts or elements (board) when other different behavior is detected
• What the shortest path is between a Pac-man/Ms.Pac-man and ghosts or other elements (board layout walls) that should have determine a different outcome?
• The ghosts’ states (chase, scatter, frightened) and modes
Special zones: Pac-Man does not allow the ghosts to turn up in the red zones (see Pac-man Dossier - Part 1) when they are in chase or scatter state. Because of that, they can only crossing the zone from right-to-left or left-to-right. Ms. Pac-man does not have those type of zones in different ways.
No comparison between Pac-man ghost and Ms. Pac-man ghost behaviour and AI is complete without not yet take into considerations a few key points mentioned here baove and also in Part 1 and Part 2. The aspects covered so far clearly do show us that behaviour differences are more to it than just the basic calculation of .. "Is the ghost in question (any one close to pac-man/mspacman character) "4 or more tiles closer or further from the determine tile" or "4 or less tiles from the determined tile" ? Will comment further on this thought in Part 5/6