Συντάχθηκε 07-10-2014 14:15
από Esthir Gelasaki
Email συντάκτη: egelasaki<στο>tuc.gr
Ενημερώθηκε:
-
Ιδιότητα: υπάλληλος.
ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ
Σχολή Ηλεκτρονικών Μηχανικών και Μηχανικών Υπολογιστών
Πρόγραμμα Προπτυχιακών Σπουδών
ΠΑΡΟΥΣΙΑΣΗ ΔΙΠΛΩΜΑΤΙΚΗΣ ΕΡΓΑΣΙΑΣ
ΔΗΜΗΤΡΙΟΥ ΤΡΙΓΚΑΚΗ
με θέμα
Σχεδίαση και Υλοποίηση ενός Αυτόνομου Πράκτορα
για το Παιχνίδι “League of Legends”
Design and Implementation of an Autonomous Agent for the “League of Legends” Game
Πέμπτη 9 Οκτωβρίου 2014, 9πμ
Αίθουσα Εργαστηρίου Intelligence, Κτίριο Επιστημών, Πολυτεχνειούπολη
Εξεταστική Επιτροπή
Αναπληρωτής Καθηγητής Μιχαήλ Γ. Λαγουδάκης (επιβλέπων)
Επίκουρος Καθηγητής Γεώργιος Χαλκιαδάκης
Καθηγητής Μιχάλης Ζερβάκης
Abstract
After the commercial success of the video game “Dota”, there has been increasing attention given to the Multiplayer online battle arena (MOBA) subgenre of Real Time Strategy (RTS) games. The creation of agents able to play autonomously within such games is sometimes limited by the absence of a public application programming interface (API). This applies to the popular game “League of Legends”, which was greatly inspired by Dota. The few computer-assisted players provided by the designers of this game range from beginner to intermediate, but have direct access to the private API inside the game. This thesis introduces a novel way to handle autonomous agent creation in such games, where access to the game state is limited to the information displayed on the user’s screen. The proposed methods come close to what a human player does, since there is a perception phase, which relies mainly on visual analysis, and a decision phase, whose outcome affects the game through emulation of the keyboard and mouse input devices. To achieve this we use screen capture on the game’s interface and computer vision algorithms to detect important information. Then, we use artificial intelligence algorithms to encode behaviors for the game character we control. Realizing this perception-decision-action cycle is very demanding in terms of computational resources, however our optimized implementation manages to meet the real-time requirements of the game. Our autonomous agent for the “League of Legends” game is able to achieve intermediate level of play and is quite competent against the designer-provided agents and also against beginner human players.
