Cristin-resultat-ID: 1986883
Sist endret: 20. januar 2022, 20:57
Resultat
Mastergradsoppgave
2021

Autonomous ship maneuvering with guaranteed safety

Bidragsytere:
  • Nora Åsheim

Utgiver/serie

Utgiver

Norges teknisk-naturvitenskapelige universitet
NVI-nivå 0

Om resultatet

Mastergradsoppgave
Publiseringsår: 2021
Antall sider: 148

Klassifisering

Vitenskapsdisipliner

Skipsteknologi

Emneord

Dynamisk Posisjonering • Fartøystyring • Marin kybernetikk

Fagfelt (NPI)

Fagfelt: Konstruksjonsfag
- Fagområde: Realfag og teknologi

Beskrivelse Beskrivelse

Tittel

Autonomous ship maneuvering with guaranteed safety

Sammendrag

During a voyage, autonomous ships usually follow a nominal path constructed prior to the voyage. Known, static obstacles can be avoided in the planning process, but unknown or dynamic obstacles cannot. Hence, such obstacles may force the vessel to deviate from the nominal path, and typically return to the path after an evasive maneuver is performed. However, it may be more advantageous to re-plan the desired path, particularly in situations with multiple obstacles. This could eliminate the need for several evasive maneuvers. By implementing a two-dimensional path variable with an along-path distance and a path normal distance, the desired vessel position can be interpreted as a virtual planar vessel. By employing control barrier functions to restrict the evolution of the path variable, the path traced out by the virtual vessel can be ensured safe. Yet, safety is only ensured for the autonomous ship if it is able to follow the path. Therefore, the feasibility of such paths must be investigated. A comparison of various methods is of value to determine which provides a feasible path or the most feasible path. A system consisting of a virtual vessel at guidance level, along with a line-of-sight guidance law was created. Three virtual vessel models from literature have been implemented. These were a first order particle model, a unicycle model, and a potential function. The virtual vessels' safety-critical controllers employed barrier functions, referred to as control barrier functions (CBFs), and quadratic programming to produce safe control inputs to the virtual vessel models. The simulations were performed in MATLAB, and thus the MATLAB function "quadprog" was used to perform quadratic programming optimization. Different CBFs from literature were implemented depending on the virtual vessel model. These included both hybrid and non-hybrid CBFs. The line-of-sight guidance produced a course angle reference for the autonomous ship based on the cross-track error to the position of the virtual vessel. The autonomous ship should perform trajectory tracking of the virtual vessel. A maneuvering model for an idealized ship was therefore implemented. The control system consisted of a backstepping controller. A control allocation was outside the scope of the thesis. The results showed a varying degree of feasibility of the paths traced out. The most promising result was produced by the unicycle virtual vessel model, with a non-hybrid CBF and a modified obstacle in the safety-critical controller. This conclusion is largely based on the along- and cross-track errors observed, and the magnitude and smoothness of the control inputs required to perform trajectory tracking. The unicycle model with a hybrid CBF performed fairly well, but experienced some switching back and forth between CBFs during the evasive maneuver. The switching between CBFs caused undesirable transient response in the system. The particle model, which employed linear velocities as control inputs, resulted in a path with sharp turns, that the ship struggled to track to a certain degree. This resulted in sub-optimal behavior, that may not be realistic, although the ship appeared to be able to track the virtual vessel during the simulations. The potential function required unrealistically high forces and moment for the ship to perform trajectory tracking. Hence, the path traced out was deemed infeasible. Several improvements are discussed. Examples are further tuning of individual parameters in the different methods, and the inclusion of a control allocation in the control system. In addition, an extension into path following, and the addition of several, and potentially dynamic, obstacles are of interest in future work. Furthermore, only current forces were added in the simulations. Including more environmental forces is an important extension of the environment to make it more realistic. However, such an extension would require the inclusion of an observer, which was outside

Bidragsytere

Nora Åsheim

  • Tilknyttet:
    Forfatter
Aktiv cristin-person

Roger Skjetne

  • Tilknyttet:
    Veileder
    ved Institutt for marin teknikk ved Norges teknisk-naturvitenskapelige universitet

Mathias Huuse Marley

Bidragsyterens navn vises på dette resultatet som Mathias Marley
  • Tilknyttet:
    Veileder
    ved Institutt for marin teknikk ved Norges teknisk-naturvitenskapelige universitet
1 - 3 av 3