Navy submarine commanded by artificial intelligence


HMS Anson Naming Ceremony

HMS Anson Naming Ceremony at Barrow-in-Furness

On April 20, the Royal Navy’s latest nuclear hunter killer submarine, HMS Anson, emerged from the vast construction hall of Barrow-in-Furness and went down the slipway into the water. All 7,400 tons of it.

Another submarine debuted on the same day in Plymouth, about 260 miles away. This secret 9-ton ship, a minnow compared to HMS Anson, could have a greater impact on the future of the Navy than a £ 1.3 billion nuclear ship.

Plymouth’s MSubs, an autonomous underwater vehicle specialist, has signed a contract with the Department of Defense to manufacture and test a super-sized unmanned underwater vehicle (XLUUV) that can operate up to 3,000 miles from home for three months.

The big innovation here is autonomy. The movement and behavior of the submarine is completely controlled by artificial intelligence (AI).

HCl Thompson is a recent graduate studying for a Master’s degree in Robotics from the University of Plymouth. He also works at Marine AI, an MSubs arm that fits the brain of XLUUV.

Mr. Thompson has no doubt about the challenges he and his colleagues are facing. “I know that many people aren’t confident in AI, so use testable elements to break things down into boxes.”

He divides AI issues into components and mission management is the most difficult. It attempts to simulate the presence of a captain trained in programming a small submarine.

This means that AI is functioning completely isolated from human contact. Especially because it is important to maintain a strict radio prayer for the submarine’s secret role. The technical principle here is machine learning, which shows an example of an AI program on how to perform a task until the appropriate action is incorporated into its own repertoire.

To do this, MarineAI uses a huge IBM AC922 supercomputer. This is Thompson’s proud “one of the biggest monsters in South West England.” In contrast, the submarine’s onboard brain is housed in a 15 cm square box, relying on the Nvidia chip commonly found in driving computer games.

“We built it from the back of the Nvidia chip because it’s energy efficient,” says Thompson. This approach consumes very little power, much like teaching school children basic programming tasks using a small but powerful Raspberry Pi device. It is also important to minimize power consumption in order to prolong the life of a submarine’s battery.

It is clear that the Ministry of Defense is paying close attention to the battery technology of this project. So far, Marine AI relies on existing technology derived from car batteries. However, he states that research in this area is making great strides.

Computer AI content needs to prioritize tasks. The project envisions a boat that can travel to submarine areas to search for land mines, plant advanced electronic intelligence packages, and stay there to explore the environment for information about enemy navies. I will.

That’s why MarineAI creates the decision-making power of the submarine’s brain. Find out how much battery life is left and how to weigh it against common weather conditions and sea conditions, making a logical decision as to whether to sail or return in the face of strong tides. To reach.

The sea that crosses remains in this project. For example, how can a submarine detect small objects on the surface, such as jet skis?

All of these dilemmas are small beers for the captain of a human submarine. Commander Ryan Ramsey was captain of the hunter killer HMS Turbulent and taught Perisher, a five-month course used by the Navy to push submarine skipper candidates to the limit.

Commander Ryan Ramsey

Ryan Ramsey was captain of the hunter killer submarine HMS Turbolent

This determines if they are suitable for commanding more than 100 submarines operating a punishment clock consisting of 6 hours on and 6 hours off during a long patrol.

Cdr Ramsey has been working with AI after leaving the Navy to see where smart software and human instincts may not produce the same response.

“AI has a hard time matching human decision-making skills. There are many submarine skills that can be transitioned to AI, but we have to accept that the first generation is not perfect.”

He studied the hostile NATO submarine commander in the exercise and learned how each individual was ready to push the ship. “In reality, if a man doesn’t know how to operate, he will find your submarine before you find him. You can’t duplicate that emotional input with AI.”

More technology in business

When XLUUV is armed, serious problems can occur. The decision to launch a torpedo is based on the instincts and vast reserves of experience of each officer. “Leaving it to a rules-based system can escalate things. There’s a lot to learn about underwater battlefields.”

In the future, Cdr Ramsey believes that submarines will have the ability to launch their own autonomous vessels. The next generation of commissioned surface vessels are already planned to be equipped with a mission bay for launching unmanned spacecraft. Autonomous submarines can also be deployed by transport, allowing battery rest and the spread of XLUUV to the world.

The long-term outlook for AI-controlled submarines is rosy, says Cdr Ramsey. “I can see how to do Perisher for AI. Let the right people recreate their experience in the simulator and let the AI ​​learn from it, including human decisions that may be wrong. . Give me 10 years worth of data and you will be able to build your own rules. “

This simulator-based computer code course promises another big advantage. “AI can test its own rules without risk. My concern about autonomous submarines is the political implications of a vessel doing something wrong or getting caught in hostile waters.” He says.

Boeing Echo Ranger Illustration

Boeing is working on a U.S. Navy autonomous submarine

The Plymouth test submarine is based on the MSubs design known as the S201. And across the Atlantic, Boeing is building a series of large unmanned submersibles as the US Navy is considering how to use these vessels.

The former submarine talks about the degree of teamwork and close comrades needed to carry out the mission. The AI ​​can’t reproduce this, but it’s not necessary if it’s doing something to save the sailors from the exhausted and repetitive tasks properly assigned by the drone.

The Royal Navy is struggling with plans for this technology and is working on a next-generation manned nuclear submarine. Still, the work of Plymouth’s AI pioneers may see their most dangerous missions delegated to an AI program that is never overlooked.