Decision-making while interacting with unmanned vessels
Authors: Vítor Fernando Plácido Da Conceição, Beatriz Sousa, Pedro Água, Joakim Dahlman
Abstract: The presence of autonomous vehicles in the maritime domain is already a reality, even though being confined to very specific domains of operations (environmental monitoring, surveillance and defense, R&D) or segregated spaces (exclusive spaces for the operation of autonomous vehicles). Artificial Intelligence algorithms for navigation control applied in autonomous vessels are based on the adoption of rules that currently regulate navigation, namely the International Collision Regulation (ColReg), the maritime Buoyage System, and routing regulations. However, considering Jen Rasmussen's decision model, in many situations, the navigator makes decisions not only based on rules (Rule-Based) but based on perceptions that stem from his skills (Skill Based) or knowledge (Knowledge-Based). An example is the concept of safe speed or safe distance, defined in ColReg, but with a variable quantification depending on the circumstances. On the other hand, the navigator's perception of the concept of navigation safety varies significantly and usually goes beyond the ship domain. For instance, some may decide not complying a ColReg priority rule to facilitate another vessel's movement and prevent a decrease in the operation safety level. Safety perception is conceived holistically, that is, it is not restricted to the vessel, but to all those in the vicinity and the natural environment. Finally, it is important to understand the behaviour of navigators in the face of the existence or interaction with unmanned vessels, not only to understand how the decision process is affected but also to improve the AI algorithms applied in autonomous vehicles.To understand how the perceived status of the encountered vessels affects the navigator's decision, we conducted an experimental study to assess how the decisions made by the participant vary when interacting with unmanned vessels. Recognizing that trust in automation is a critical influential factor, we adopted existing framework models to evaluate the participants' perceptions of Maritime Autonomous Surface Ships (MASS), as classified by the International Maritime Organization.The adopted method comprises a combination of questionnaires and participation in six simulated scenarios. This mixed approach aimed to understand the familiarity with MASS; the need to change operational regulations; concerns, challenges, and opportunities from the implementation of MASS; trust in MASS; and the differences between the declared perception and decision-making when interacting with MASS.The study comprised three stages. firstly, a pilot study to appraise and validate the questionnaire, with 49 participants. Secondly, the online implementation of the questionnaire, with a desktop version of the six simulated scenarios, with 110 valid questionnaires, 73 students from the naval academy and 37 professional mariners. Each scenario presented an interaction situation with another vessel, referencing a clearly stated rule of the Collision Regulation. The target vessel could randomly assume one of three statuses: Manned vessel, Unmanned vessel and unidentified vessel. By varying the control mode of the target vessel in the same situation, we aimed to see if the perceived status of the vessel had any influence on the decision-making process. In the last stage of the study, the six desktop exercises of the scenarios were replaced by a simulator game of the same situation, with 33 participants. On the desktop exercise participants reported: Time for acting, change of heading, change of speed, and aimed final position. Reaction time, change of heading and speed were automatically logged on the simulator game. The questionnaire comprises four sections: Unmanned vessels and levels of automation, scenarios decisions, trust in automation and demographic data.The results suggest that despite having a reduced familiarity with autonomous ships, the participants have a very positive opinion. However, in the same situation, they react differently to conventional ships and autonomous ships. The way navigators react was analyzed through parameters such as reaction time, course and speed variation and the Closest Point of Approach between vessels. There is a greater discrepancy between those parameters in participants with less training, suggesting a need to address the issues of interaction with unmanned vessels during the course program. Results from the simulators provided more precise shreds of evidence, namely when interacting with unidentified vessels, pointing out the need to design solutions for clear identification of the target vessel.
Keywords: Trust in automation, MASS, Autonomous Ships, Safety, Decision-making
Cite this paper: