Umetno življenje na primeru kolonije inteligentnih agentov

RITLOP, ROK (author), Robnik Šikonja, Marko (mentor)

Izdelali smo simulacijo umetnega življenja na primeru kolonije inteligentnih posameznikov (agentov), pri čemer smo uporabili kombinacijo nevronskih mrež, genetskih algoritmov in optimizacije s kolonijo mravelj. Vsak agent ima svoje možgane, implementirane z nevronsko mrežo. Večje število agentov tvori kolonijo, ki ima interakcijo z okoljem preko nabiranja in shranjevanja hrane, med seboj pa komunicirajo s pomočjo feromonskih sledi. Preko dedovanja in mutacije že obstoječih nevronskih mrež je omogočen razvoj kolonije. S simulacijo smo pridobili podatke o učinkovitosti kolonij z različnimi stopnjami mutacije (0,01 do 0,10) in rezultate med seboj primerjali. Kolonije z nizkimi stopnjami mutacije so bile v splošnem manj uspešne, kolonije z visokimi stopnjami mutacije pa so uspešno in hitro razvile le nekatera vedenja. Za najbolj uspešne so se izkazale kolonije s stopnjo mutacije 0,05, ki predstavlja dobro ravnovesje med kreiranjem novih agentov in kopiranjem že obstoječih uspešnih agentov. Simulacija omogoča spreminjanje parametrov in predstavlja dobro osnovo za nadaljnji razvoj in nadgradnje ter dodajanje bolj kompleksnih vedenj agentov.

• Language: Slovenian
• Keywords: umetno življenje, nevronske mreže, genetski algoritmi, optimizacija s kolonijo mravelj
• Work type: Undergraduate thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2015
• PID: 20.500.12556/RUL-30856
• Publication date in RUL: 02.07.2015

RITLOP, ROK, 2015, Umetno življenje na primeru kolonije inteligentnih agentov [online]. Bachelor’s thesis. [Accessed 31 March 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=30856

Secondary language

• Language: English
• Title: Artificial life on a colony of intelligent agents

Abstract

We developed a simulation of artificial life with a colony of intelligent individuals (agents) by using a combination of neural networks, genetic algorithms, and ant colony optimization. Each agent has its own brain implemented as a neural network. Several agents form a colony which interacts with the environment by gathering and storing food. They communicate using pheromone trails. Through the processes of inheritance and mutation of agents' brain the colony can develop continuously. With simulation we gathered the information on the effectiveness of colonies with varying rates of mutation (from 0.01 to 0.10) and compared the results. The colonies with low mutation rates were overall less successful, while the colonies with high mutation rates were successful in developing only certain behaviours. The most successful colonies used the mutation rate of 0.05, which presents a good balance between creating new agents and copying the existing successful agents. The simulation allows adjustment of parameters and presents a good basis for further development and adding more complex agent behaviours.

• Keywords: artificial life, neural networks, genetic algorithms, ant colony optimization