University of Twente Student Theses


Using reinforcement learning to control hydrofoils

Schaaf, Jurre van der (2022) Using reinforcement learning to control hydrofoils.

[img] PDF
Abstract:Hydrofoils on vessels are used to lift a vessel's hull above the water surface to reduce drag and increase efficiency and top speed. By changing the angles of the hydrofoils, more or less lift can be created. Controlling the different foil angles makes it possible to regulate the height and stability of the ship. However, these control systems can not always adapt to a new situation when the environment changes or when it is placed in a new environment, leading to a malfunctioning system. This research investigates the approach of a control system based on reinforcement learning. This is done by implementing Deep Q-Learning on a simulation of a hydrofoil boat. For this, different configurations have been tested and compared against a PID controller. These configuration differ in actions sets, reward functions and number of simultaneous agents. Examination of plots, as well as the standard deviation of the roll and pitch angles of the vessel, showed that using a dedicated agent per foil with a small action set with a reward based on height, the performance is comparable to a PID controller and considered stable. Other experiments with larger actions sets, one agent controlling two foils simultaneously or rewarding by roll angle showed that the system was in those cases not able to perform its task properly.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page