PhD defence: Digital Cognitive Companions for Marine Vessels - On the Path Towards Autonomous Ships
Plats: Online at the zoom platform (Link by registration)
Kontakt: per [dot] runeson [at] cs [dot] lth [dot] se
Spara händelsen till din kalender
Thesis title: Digital Cognitive Companions for Marine Vessels - On the Path Towards Autonomous Ships
Author: Mårten Lager, Department of Computer Science, Lund university Wallenberg AI, Autonomous Systems and Software Program (WASP).
Faculty opponent: Prof. Henrik I. Christensen, University of California, USA
Location: Online - link by registration
- Improving conditions for autonomous ships. Interview with Mårten Lager at lth.se or ai.lu.se
- The thesis for download at portal.research.lu.se.
As for the automotive industry, industry and academia are making extensive efforts to create autonomous ships. The solutions for this are very technology-intense. Many building blocks, often relying on AI technology, need to work together to create a complete system that is safe and reliable to use. Even when the ships are fully unmanned, humans are still foreseen to guide the ships when unknown situations arise. This will be done through teleoperation systems.
In this thesis, methods are presented to enhance the capability of two building blocks that are important for autonomous ships; a positioning system, and a system for teleoperation.
The positioning system has been constructed to not rely on the Global Positioning System (GPS), as this system can be jammed or spoofed. Instead, it uses Bayesian calculations to compare the bottom depth and magnetic field measurements with known sea charts and magnetic field maps, in order to estimate the position. State-of-the-art techniques for this method typically use high-resolution maps. The problem is that there are hardly any high-resolution terrain maps available in the world. Hence we present a method using standard sea-charts. We compensate for the lower accuracy by using other domains, such as magnetic field intensity and bearings to landmarks. Using data from a field trial, we showed that the fusion method using multiple domains was more robust than using only one domain.
In the second building block, we first investigated how 3D and VR approaches could support the remote operation of unmanned ships with a data connection with low throughput, by comparing respective graphical user interfaces (GUI) with a Baseline GUI following the currently applied interfaces in such contexts. Our findings show that both the 3D and VR approaches outperform the traditional approach significantly. We found the 3D GUI and VR GUI users to be better at reacting to potentially dangerous situations than the Baseline GUI users, and they could keep track of the surroundings more accurately. Building from this, we conducted a teleoperation user study using real-world data from a field-trial in the archipelago, where the users should assist the positioning system with bearings to landmarks. The users experienced the tool to give a good overview, and despite the connection with the low throughput, they managed through the GUI to significantly improve the positioning accuracy.
Please register at www.lth.se/digitalth/events/register-2021-02-12/ in order to get an access link at the zoom platform.