Aprendizaje automático aplicado a juegos de estrategia en tiempo real: un enfoque genético
Loading...
Official URL
Full text at PDC
Publication date
2020
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
La aplicación de técnicas de inteligencia artificial a juegos de estrategia en tiempo real es un campo de investigación amplio con numerosos problemas abiertos. Para probar estas técnicas antes de aplicarlas a dominios complejos, se creó el juego µ-RTS, una versión simplificada del género de estrategia en tiempo real. Nosotros proponemos el uso de comportamientos precodificados para cada tipo de unidad del juego, que se seleccionan formando estrategias globales que determinan las decisiones de la IA durante una partida. En un paso mayor de complejidad, utilizamos varias estrategias durante la misma partida, lo que permite a la IA cambiar su comportamiento en función de la etapa de la partida en la que se encuentre. Para confeccionar las estrategias y seleccionar las más adecuadas para una partida, utilizamos un algoritmo genético. Finalmente, incluimos un análisis de cómo se desenvuelve nuestro planteamiento frente a otros bots desarrollados para el juego.
The use of artificial intelligence techniques to play real-time strategy games is a broad field of research with many open problems. The game µ-RTS, which is a simplified version of an RTS game, was created to test this techniques before using them in more complex domains. We present a bot that uses hard-coded behaviors designed for each different type of unit in µ-RTS. Some of these behaviors are selected to create a strategy that the bot can use during the game. Moreover, several strategies can be used in the same game, which lets the bot change its way to play depending on the stage of the game. We use a genetic algorithm to create these strategies and choose the best ones for a particular game. Finally, we include a study about how this approach performs against other state-of-the-art bots created for µ-RTS. This document is mainly written in Spanish. However, a brief introduction in English can be found at page 5, and some conclusions of the work can be found at page 79.
The use of artificial intelligence techniques to play real-time strategy games is a broad field of research with many open problems. The game µ-RTS, which is a simplified version of an RTS game, was created to test this techniques before using them in more complex domains. We present a bot that uses hard-coded behaviors designed for each different type of unit in µ-RTS. Some of these behaviors are selected to create a strategy that the bot can use during the game. Moreover, several strategies can be used in the same game, which lets the bot change its way to play depending on the stage of the game. We use a genetic algorithm to create these strategies and choose the best ones for a particular game. Finally, we include a study about how this approach performs against other state-of-the-art bots created for µ-RTS. This document is mainly written in Spanish. However, a brief introduction in English can be found at page 5, and some conclusions of the work can be found at page 79.
Description
Trabajo de Fin de Grado en Ingeniería Informática y Matemáticas, Facultad de Informática UCM, Departamento de Ingeniería del Software e Inteligencia Artificial, Curso 2019/2020
En esta memoria se describe el desarrollo de una inteligencia artificial capaz de jugar a un juego de estrategia en tiempo real. El código correspondiente puede encontrarse en el repositorio de GitHub https://github.com/TFG-Informatica/ Aprendizaje-automatico-aplicado-a-juegos-RTS