On September 16, 1620, a group of English Puritans gathered their belongings and set sail for what they viewed as the New World. Despite treacherous weather, the captain of the Mayflower ensured their safety and almost all passengers survived the two-month journey to what is now known as Plymouth, Massachusetts. 

A ship’s captain bears a heavy responsibility, and until recently this role has been reserved solely for humans. However, the Mayflower Autonomous Ship (MAS) might help change that.

The ship, which is meant to commemorate the 400^th anniversary of the Mayflower’s voyage, is the product of collaboration between IBM ProMare and the Marine AI section of the Submergence Group. We spoke with Don Scott, director of engineering for Submergence Group, to get some insight into the unique challenges behind creating an autonomous ship.

Not a car

For the most part, discussion about autonomous transportation has been dominated by self-driving cars. However, ships at sea present a nearly ideal environment for a robot navigator. Once out of port, there aren’t many obstacles to run into in the open ocean. Still, there are challenges the team needed to consider.

“Self-driving cars are very similar to what we're doing but the challenges are quite different,” says Scott. “We’re typically working at lower speeds, so we don't need the crazy update rates that they're dealing with. But our environment is incredibly dynamic. Our landscape is constantly moving and shifting and impacting our ship.”

While autonomous cars need to make decisions in milliseconds, that’s rarely necessary on the open seas. The MAS evaluates its environment once every second, and Scott says that’s plenty. 

That said, collisions do occur on the ocean, and there are strict protocols for how to react in any situation. To make sure everyone knows what they’re doing, maritime experts have developed a set of rules called Collisions Regulations, or COLREGs.

“Under COLREGs, if a ship is approaching me off my port, he has to give way and understand that so I can continue,” says Scott. “But sometimes people don't follow COLREGs. So that’s where the reinforcement learning comes in. This interaction is a little bit of a dance. And yes, it's highly regulated, but you do need to be able to respond to what the other ships doing.”

Here, Scott is talking about the machine learning process that makes it possible for the MAS to tell the difference between a ship following COLREGs and one that may become a threat. As in a dance, the ship has to be able to read her “partner’s” actions to predict what they’ll do next.

The team used an IBM AC922 with four GPUs to build the models needed to identify objects while at sea. Then, the team will deploy them on Nvidia Jetson AGX Xavier devices on the ship for object detection, classification, and tracking.

Sailing on the edge

While the machine learning technology is interesting enough, the truly remarkable aspect of this project has to do with the edge computing it will require to function. Edge computing is where certain computational functions are performed at the ‘edge’ of the network. In this instance, that means the MAS will both capture data and process it right on the ship without having to send it to another computer on the mainland.

“Previously, I’ve called the Mayflower the ultimate edge device,” says Scott. “But it's really an edge system made up of edge systems. All the cameras are intelligent edge devices. Even the battery management system is an intelligent edge device.”

Edge computing will help the ship assess dangers and understand how to avoid them while at sea. However, the most important idea is to have a ship that is able to handle itself autonomously for long periods. 

“One of the original mandates of the ship was as a research vessel for ocean science,” says Scott. “We'd love to see a fleet of these ships being used as open platforms for ocean research. It’s very expensive to collect oceanographic data. Autonomous vessels provide a flexible, cost-effective option for scientists to help gather data. The vision is for autonomous ships working in tandem with humans and other autonomous vessels to go further, for longer, and capture more comprehensive data than currently possible.”

Sadly, the MAS wasn’t able to set sail on the 400^th anniversary of the Mayflower. The pandemic played a role in the delay, but Scott said poor weather may have kept the ship in port anyway. However, the team did get to partake in a naming ceremony in Plymouth, England and Scott expects the ship to set sail sometime in spring 2021.

That said, this project wasn’t really about the Mayflower. If anything, this is more about the underlying mission of the Mayflower rather than its exact journey—setting sail into the unknown with hopes for building a better world.

“This is why I became an engineer,” says Scott. “I've had a long career in marine engineering  and ocean engineering. And I think what I like the best are these big collaborative projects. This is a hard project, right? It’s not easy! We’re trying some pretty new things and we're making mistakes, but we're figuring it out. It's actually the most fun I've ever had on a project.”