| dc.description.abstract | The increased autonomy in maritime shipping is expected to make the industry safer, more efficient, and reduce its environmental impact. Still, efficient interactions between autonomous agents and humans are fundamental to increasing trust and reducing the risk of collision. One challenge for Maritime Autonomous Surface Ships (MASS) is facilitating interactions through radio between ships and shore stations. Despite the importance of VHF for safe maritime traffic, current MASS prototypes are not able to understand nor participate in this communication. This thesis proposes using a conversational user interface (CI) with the autonomous agent in MASS as a replacement for radio in vessel-to-vessel and vessel-to-infrastructure communication. Based on maritime communication guidelines and textual descriptions of a MASS’ surroundings, we demonstrate the use of the transformer-based deep learning model GPT-3 to enable human-like answers to prompts of status, decisions, and future intent. Usability tests with ship officers and Vessel Traffic Service operators (N=9) suggest that in peer-to-peer conversations, human operators are able to obtain a perceived sense of situational awareness from the MASS through the CI. A discussion of the opportunities and limitations of CIs in a maritime shipping environment saturated by autonomous systems concludes the thesis. | |
