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


Jetpack Attack: AI for Project Hoshimi

Authors

;

Term

4. term

Education

Publication year

2008

Abstract

Dette speciale beskriver, hvordan vi byggede Jetpack Attack, en AI til programmeringsspillet Project Hoshimi. Arbejdet undersøger, hvordan man implementerer en komplet multi‑agent‑AI—flere softwareagenter der samarbejder—men fokuserer på to praktiske valg, som har stor betydning: at finde et godt start-/landingspunkt og at klassificere missionen. Først gennemgår vi spillets regler og den overordnede systemudformning. Dernæst gennemgår vi en række AI‑teknikker, implementerer dem der viser sig nyttige, og udvikler en ny metode til at finde landingspunktet. Til sidst evaluerer vi systemet gennem kontrollerede eksperimenter og ved at indsende det til en verdensomspændende konkurrence, præsenterer resultaterne og vurderer, hvad der virkede og hvad der ikke gjorde.

This thesis documents how we built Jetpack Attack, an AI for the programming game Project Hoshimi. The work explores implementing a complete multi‑agent AI—multiple software agents that coordinate—but focuses on two practical decisions that strongly affect play: choosing a good starting (landing) point and classifying the mission. We first outline the game rules and the overall design of the AI. We then review a range of AI techniques, implement those that prove useful, and develop a new method for selecting the landing point. Finally, we evaluate the system through controlled experiments and by submitting it to a worldwide competition, report the results, and assess what worked and what did not.

[This abstract was generated with the help of AI]