AAU Student Projects - visit Aalborg University's student projects portal
A master's thesis from Aalborg University
Book cover


Jetpack Attack: Kunstig Intelligens til Project Hoshimi

Translated title

Jetpack Attack: AI for Project Hoshimi

Authors

;

Term

4. term

Education

Publication year

2008

Pages

96

Abstract

Denne afhandling beskriver arbejdet med at bygge en konkurrencedygtig AI, Jetpack Attack, til programmeringsspillet Project Hoshimi. Målet er at implementere en komplet multi‑agent AI—software der koordinerer flere autonome enheder—men fokus er på to opgaver: at vælge et indledende landings-/startpunkt og at klassificere missionen, så AI’en kan vælge passende strategier. Først gennemgår vi reglerne i Project Hoshimi og det overordnede design af vores AI. Dernæst gennemgår vi en række teknikker inden for maskinintelligens, implementerer dem der viser sig nyttige, og introducerer en ny metode til at finde landingspunktet. Den færdige løsning evalueres gennem kontrollerede eksperimenter og ved at deltage i en verdensomspændende konkurrence. Vi præsenterer resultaterne af disse tests og afslutter med en vurdering af, hvad der fungerede, hvad der ikke gjorde, og hvordan de valgte teknikker påvirkede ydeevnen.

This thesis reports on building a competitive AI, Jetpack Attack, for the programming game Project Hoshimi. The goal is to implement a complete multi‑agent AI—software that coordinates several autonomous units—but with a focus on two tasks: choosing an initial landing/start point and classifying the mission so the AI can select suitable strategies. We first outline the rules of Project Hoshimi and the overall design of our AI. We then review a range of machine intelligence techniques, implement those that proved useful, and introduce a new method for selecting the landing point. The finished system is evaluated through controlled experiments and by entering a worldwide competition. We present the results of these tests and conclude with an assessment of what worked well, what did not, and how the chosen techniques affected performance.

[This abstract was generated with the help of AI]