UST 031

10 Platform one IBM and marine research organisation Promare are testing a machine learning framework for full-sized marine autonomous systems (writes Nick Flaherty). The AI Captain will be used by the Mayflower Autonomous Ship (MAS) as the first vessel to self-navigate across the Atlantic, in September. The software is being tested on a manned research vessel off the coast of Plymouth in the English Channel to evaluate how the framework uses data from sensors. The first stage tests the input from radar, AIS, GPS and navigation systems, as well as data about visibility. Cameras, computer vision, edge and autonomy capabilities will be added in the next phase of testing. The Mayflower team has been training the ship’s AI models over the past two years, with more than a million nautical images collected from cameras in Plymouth Sound as well as open source databases. The training framework runs on an IBM Power AC922 system board, using IBM Power9 CPUs and Nvidia V100 Tensor Core GPUs. The trained computer vision algorithms running on three Nvidia Jetson boards on the MAS trimaran should be able to independently detect and classify ships, buoys and other hazards such as land, breakwaters and debris. The Jetson boards use the Xavier GPU with multiple ARM processors, so they are different from the IBM Power processors that will be running Red Hat Enterprise Linux (RHEL) as the system controller. Some of the main restrictions to the use of autonomous shipping are the marine regulations. Alongside the vision processing algorithms running on the Jetson boards, the controller will use IBM’s rule management system, called the Operational Decision Manager. This implements the International Regulations for Preventing Collisions at Sea as well as recommendations from the International Convention for the Safety of Life at Sea. This provides a completely transparent record of its decision-making process, avoiding the problems of a machine learning framework where the decisions are not visible. A Safety Manager function running on RHEL will review all the AI Captain’s decisions to ensure they are safe, for the Mayflower as well as other vessels in its vicinity. “Many of today’s autonomous ships are really just automated – they are robots that do not adapt dynamically to new situations and rely heavily on operator override,” said Don Scott, CTO of the MAS project. “Using an integrated set of IBM’s AI, cloud and edge technologies, we are aiming to give the Mayflower the ability to operate independently in some of the most challenging places on the planet.” Captain AI coming aboard Marine vessels April/May 2020 | Unmanned Systems Technology A machine learning framework for the Mayflower autonomous trimaran is currently being tested at sea