Vodenje psa ovčarja z uporabo umetne inteligence

Marinko, Anže (author), Lebar Bajec, Iztok (mentor), Demšar, Jure (comentor)

V naravi lahko en sam ovčar zbere in vodi čredo več sto ovc, ljudje pa si želimo zbirati, voditi ali preusmerjati tudi druge vrste živali. V bližini letališč imajo letala pogosto težave z jatami ptic, ki letalom med drugim povzročajo škodo na motorjih ali celo strmoglavljanja. V ta namen so že mnogi raziskovali različne modele vodenja s pomočjo umetne inteligence, da bi se podobnih problemov lahko lotili z roboti ali droni. Uporaba pametnih ovčarjev bi lahko prav prišla tudi pri zbiranju razlite nafte na vodi ali pri vodenju panične množice ljudi na varno v kriznih situacijah. V tem delu študiramo različne modele gibanja ovc, razvijamo model vodenja in ga še izboljšamo z metodami umetne inteligence. Pri tem predlagamo rešitve za različne težave, ki se pri učenju pojavljajo. Razviti želimo namreč model, po katerem ovčarji kar se da hitro v stajo prepeljali celotno čredo. Osnovnemu modelu vodenja najprej dodamo sodelovanje več ovčarjev, nato najdemo dobre parametre modela za izbrane velikosti črede in število ovčarjev. Na koncu razvijemo model, pri katerem se na podlagi izbranih informacij ovčar sproti odloča o vrednostih parametrov.

• Language: Slovenian
• Keywords: umetna inteligenca, problem vodenja ovčarja, sodelovanje, genetski algoritmi, spodbujevano učenje, agenti
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FMF - Faculty of Mathematics and Physics; FRI - Faculty of Computer and Information Science
• Year: 2021
• PID: 20.500.12556/RUL-126885
• COBISS.SI-ID: 62894595
• Publication date in RUL: 08.05.2021

MARINKO, Anže, 2021, Vodenje psa ovčarja z uporabo umetne inteligence [online]. Master’s thesis. [Accessed 11 April 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=126885

Secondary language

• Language: English
• Title: Guide a shepherd dog using artificial intelligence

Abstract

In the wild, a single shepherd can gather and lead a herd of hundreds of sheep, and people want to collect, lead or divert other species of animals as well. In the vicinity of airports, aircraft often have problems with flocks of birds, causing them damage to their engines or even crashes. Many have already researched various models of guidance using artificial intelligence in order to be able to tackle similar problems with robots or drones. The use of smart shepherds could also come in handy when collecting spilled oil on the water or when leading a panicked crowd of people to safety in crisis situations. In this part, we study different models of sheep movement, develop a management model and further improve it with artificial intelligence methods. We propose solutions to various problems that arise in learning. Namely, we want to develop a model according to which shepherds transport the entire herd to the stable as quickly as possible. We first add the participation of several shepherds to the basic management model, then we find good model parameters for the selected herd sizes and number of shepherds. Finally, we develop a model in which the shepherd decides on the values of parameters on the basis of selected information.

• Keywords: artificial intelligence, shepherding problem, cooperation, genetic algorithms, reinforcement learning, agents