Analiza performansi algoritma za izbjegavanja sudara na moru temeljenog na COLREG-u

master's thesis

• Title (english): Performance Analysis of COLREG-based Collision Avoidance Algorithm at Sea
• Author: Petar Stojanović
• Mentor: Nikola Mišković (mentor)
• Committee member: Nikola Mišković (član povjerenstva)
• Committee member: Đula Nađ (član povjerenstva)
• Committee member: Fausto Miguel Pascoal Ferreira (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2025-02-20, Croatia

Abstract

Ovaj rad predstavlja analizu performansi algoritma za izbjegavanje sudara u pomorskoj navigaciji, s posebnim naglaskom na metodu prepreke brzine i pristup temeljen na dubokom podržanom učenju. Istraživanje obuhvaća matematičku i geometrijsku formulaciju VO metode, njezinu implementaciju u Pythonu, integraciju sa ROS2 sustavom te simulacijske eksperimente s različitim scenarijima susreta. Rezultati pokazuju da, iako je VO metoda brza i transparentna, njena rigidnost i osjetljivost na... More nesigurnosti senzorskih podataka mogu dovesti do neočekivanih manevara. S druge strane, DRL pristup omogućuje adaptaciju u realnom vremenu bez potrebe za ručnim podešavanjem parametara, no zahtijeva opsežno treniranje. Rad također raspravlja o mogućnostima hibridnog sustava koji bi kombinirao prednosti oba pristupa uz dodatne ispravke, poput poboljšanja sustava prijenosa podataka i dosljedne konverzije koordinata. Less

Abstract (english)

This paper presents an analysis of the performance of a collision avoidance algorithm in marine navigation, with a special emphasis on the speed barrier method and an approach based on deep support learning. The research includes the mathematical and geometric formulation of the VO method, its implementation in Python, integration with the ROS2 system, and simulation experiments with different meeting scenarios. The results show that, although the VO method is fast and transparent,... More its rigidity and sensitivity to sensor data uncertainties can lead to unexpected maneuvers. On the other hand, the DRL approach enables real-time adaptation without the need for manual parameter adjustment but requires extensive training. The paper also discusses the possibilities of a hybrid system that would combine the advantages of both approaches with additional improvements, such as improving the data transfer system and consistent coordinate conversion. Less

• Keywords: COLREG; prepreka brzine; duboko podržano učenje; autonomna navigacija; izbjegavanje sudara; ROS2; simulacijski eksperimenti
• Keywords (english): COLREG; Velocity Obstacle; Deep Reinforcement Learning; autonomous navigation; collision avoidance; ROS2; simulation experiments
• Language: croatian
• URN:NBN: urn:nbn:hr:168:490080
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2025-03-12 11:18:57