The rapid advances being made with artificial intelligence (AI) and ship automation are compelling class societies to keep pace with developments and support clients keen to explore the potential advantages of these emerging technologies. But one of the difficulties for shipowners, designers, class and IMO alike is the varying degrees of autonomy. Is the technology in question truly autonomous or rather remotely operated? Does it apply to the entire vessel or just certain shipboard systems? What are the operational parameters? What contingency measures are in place?
IMO’s Maritime Safety Committee (MSC) will meet again in December to further discuss automation definitions and frameworks, with the full results of analysis expected in 2020. But irrespective, it’s clear that no single set of requirements will be adequate to govern all permutations.
Japanese classification society ClassNK arrived at the same conclusion while developing its provisional Guidelines for Concept Design of Automated Operation/Autonomous Operation of Ships, published earlier this year. Building upon the research goals it outlined in last year’s R&D Roadmap (The Naval Architect, November 2017), the guidelines detail the elements that need to be considered during the concept design development of automated ship systems, with a particular emphasis on safety.
Man and machine
The challenge is that the human-machine interface is often a blurred line when it comes to automation, and often a broad ‘task’ such as navigation is achieved through a combination of machine and human elements. ClassNK determined that it was necessary for each task to be broken into a series of ‘subtasks’. Of these subtasks, some are ‘decision making’, meaning that they draw upon external information and what would traditionally be human skills or attributes such as situational awareness.
The idea is to clearly set out the division of roles between human and machine and under what conditions responsibility transfers with each subtask, including the provision of fallback measures if something goes wrong and making clear exactly when these would take effect. Automated systems are to be defined according to their Operational Design Domain (ODD), the specific environment under which the system is intended to function.
To explain this process, ClassNK has devised a set of groupings by which all autonomous operations can be categorised (Table 1), ranging from scenarios where a limited number of subtasks are handled by automated systems (Group I), to full automation where there is no human involvement at all (Group IV). ClassNK stresses that these are only the prime examples and there may be further updates.
In developing its model, ClassNK referred to the automation philosophy being applied to automobiles and taking into consideration the characteristics that are specific to ships. It states: “As objects of automation, ships are complex and involve a variety of tasks and functions. We decided to apply the tasks/subtasks concept because it is crucial to factorise the objects of automation to the related onboard operations and works, and to have a common understanding for everyone involved with ship operations.”
The human-machine interface therefore is critical, with the risk that ambiguities may arise as automated operation systems grow more complex and their information relayed becomes more challenging to decipher.
ClassNK believes these interfaces must be based on “appropriate standards for human-centred design” that are easy to use and familiarisable. “We referred to appropriate standards for designing this kind of interface, such as ISO 9241-210:2010 (Ergonomics of human-system interaction – Part 210: Human-centred design for interactive systems),” it explains. A forthcoming set of guidelines will outline the operational requirements for automated systems and how this must correlate with procedures outlined in the vessel’s Safety Management System.
Anticipating flaws in the human-machine interface is specified as one of the necessary considerations when conducting risk assessments. Others include the impact of automated operation to the function of the overall ship, resilience of the communications network, computer systems and sensors, and any vulnerability to cyber or physical attacks.
Where some or all part of a vessel’s operations are remotely controlled, i.e. subtasks carried out from onshore, then ClassNK will exercise its discretion if it doesn’t believe the provisions made for each subtask are practical. Moreover, the guidelines expect that the person(s) in charge of the remote operation will have the same abilities as crew performing the same subtask(s). This doesn’t necessarily mean an onshore operator must be trained to full mariner standards, but with tasks such as navigation they would be expected to have the same skills and capability. The society says at this stage there have been no consultations regarding the development of training standards for remote operation but will consider this as required.
Taking a broader outlook, ClassNK says that while the past 18 months have seen a number of headline-grabbing trial projects involving smaller, remotely controlled vessels, it is actually receiving more enquiries with regard to the automation systems for larger oceangoing ships, particularly with regard to navigation technology.
It believes also that it is essential for the IACS members to exchange opinions with each other and develop “appropriate common ground on the basic perspective of verification for automated operation.”
Indeed, while the basic skeleton for an automated future is now being pieced together, the society anticipates that the role of AI systems in particular will need to come under closer scrutiny. “With all of the various onboard tasks and duties, the practical use of an ‘automated ship’ with an operation system that can specifically propose actions, and provide information to crewmembers, as the final decision-maker, will be a major challenge,” it says.
“These guidelines target the concept design of automated and autonomous operation of ships. The next stage will establish technical requirements and verification processes on all stages of automated/autonomous operation involved in the design/development, construction and entry into service, and release them as further guidelines.”