Neural Networks Attack: How AI Is Used in Yachting

Artificial intelligence is no longer a novelty, but its use in yachting isn't always obvious. Interparus will explore real-world applications of AI at sea.

The general situation with AI

The use of artificial intelligence in sailing is very different from tools based on language models. Neural networks like ChatGPT, DeepSeek, and Gemini are recent and widely publicized, but they're not really suited for work at sea. Perhaps they're useful as a quick reference or a pocket assistant. No, modern yachting is much more interesting.

Forms of artificial intelligence have existed for decades. Machine learning and neural networks trained on transaction data have been used by major banks since the mid-1990s. These technologies have significantly improved the accuracy of card fraud and other anomaly detection in real time. Since 2012, AI capabilities have rapidly expanded thanks to the development of deep learning, which has given algorithms the ability to learn from massive data sets.

Today, artificial intelligence is used in numerous areas of the maritime industry. It assists in safety systems, weather and route forecasting, ship design, performance analysis, sports coaching, and broadcasting. AI is also increasingly being used in the development of autonomous vessels, and in the coming years, such systems may begin to assist navigators with navigation and situational awareness on the bridge.

Opinions from AI industry representatives

"AI at sea doesn't have to be perfect. It just needs to perform certain tasks better than humans," says Yarden Gross, co-founder and CEO of ORCA AI, which supplies systems to shipping operators including MSC and Maran Tankers.

"We must acknowledge the obvious: the current system is unsafe. Tired crews, challenging navigation conditions, reliance on human supervision, and aging fleets with outdated equipment—all of this is not a model worth maintaining."

ORCA AI Interface

"We're already seeing that systems with AI elements provide safer and more stable watchkeeping than a tired or distracted officer, especially in storms or poor visibility," he notes. Gross also proposes a "hybrid intelligence" model, where humans and algorithms work together, each playing their own role. "A transitional format, where a vessel changes its level of autonomy depending on the situation, is the real revolution. And it's already underway," Yarden adds.

Modern ships constantly collect data, analyze near-misses, and evolve as a unified learning system. Each voyage makes them smarter, and all information from voyages feeds back into the algorithms, improving their efficiency and accuracy.

AI for Security

During the most recent Vendée Globe race, approximately 65% competitors used the Sea.ai system, which helps detect small floating objects that can cause serious damage to a yacht. Formerly known as OSCAR, this mast-mounted system combines images from a color camera with data from two thermal night vision cameras.

The onboard artificial intelligence continuously analyzes video feeds, recording any unusual changes on the ocean surface. If any deviations in temperature or image structure are detected, the system recognizes and classifies the object—whether it's a person overboard, a whale, a lost container, or another vessel—and immediately alerts the skipper.

Interest in this technology is growing rapidly: in just a few years, the company has grown from a startup into an enterprise with a team of 70 people and three offices.

Initially, the algorithms were trained using data collected during the Vendée Globe 2020–2021, but this process required a significant amount of manual work. Now, training is based on target scenarios. Some situations, such as a man overboard, are specifically modeled to improve system performance. Other cases are flagged by users themselves during their voyages.

Continuous learning on real data

"Every update makes the system smarter thanks to new data," explains Sea.ai's Solène Guerou. She explains that raw video footage and sensor data—such as man-overboard incidents or ice collisions—are regularly sent to the company's headquarters in Brittany.

Once detection is confirmed, the data is transferred to a team of specialists, who manually label each object in a way that the neural network can understand. For example, a section of the sea is highlighted in the image and a caption is added indicating that a person overboard is located in that area. This way, the system continuously learns, and the results of this learning are made available to users through software updates.

Weather Changes

Weather forecasting and route planning are another area where artificial intelligence is rapidly gaining importance. Modern weather models are based on numerical calculations and require colossal computing power, so they run on the world's most powerful supercomputers.

Weather station on the coast

However, AI can not only improve forecast accuracy but also significantly reduce energy consumption. Earlier this year, the ECMWF launched its own AI-powered forecasting system, AIFS. It operates in parallel with the traditional physical-based IFS model and produces similar results.

Learn more about how AIFS and weather agents work

Moreover, in some scenarios, such as tropical cyclone tracking, AIFS produces even more accurate forecasts while using approximately a thousand times less energy.

This reduction in resource requirements means that low-resolution global AIFS models can be run even on powerful gaming laptops. Marine Weather Intelligence, a startup founded by marine meteorology expert Christian Dumars, uses AI to optimize routes—both in ocean regattas and to reduce the carbon footprint of commercial fleets. Its clients include the Vendée Globe and The Ocean Race.

Features of AI-powered weather models

According to Dumard, the company actively uses artificial intelligence to analyze which weather models are most reliable in different conditions. This depends, for example, on the length of the route or the season. Satellites accurately measure wave heights, and AI helps identify systematic errors in wave models. "The wave forecast is fairly accurate for the first two or three days," he noted at the Yacht Racing Forum, "but over time, the bias accumulates depending on the season and other factors."

Marine Weather Intelligence Interface

AI is excellent at analyzing such complex relationships, he added. "It's difficult for humans to consider all the parameters simultaneously, but a system does it faster and more accurately." Forecasts created with AI are already being used for strategic route planning in ocean racing.

With growing interest in sail support for vessels and the transition to more cost-effective operating models, research in this area is rapidly advancing. In the long term, its results will benefit not only large operators but also ordinary yachtsmen.

Skippering and yacht management

Ocean currents are extremely difficult to accurately predict, although their eddy formations significantly affect both sea conditions and navigation times. Amphitrite, a French startup based in Paris, has proposed a solution: using artificial intelligence to analyze ocean currents based on high-resolution satellite imagery.

According to an independent study, a ship sailing at 16 knots between Tunis and Tangier was able to reduce its fuel consumption by 4% using Amphitrite data and arrive about an hour earlier.

Problems of training neural networks for control

To maintain course and account for sea conditions, autopilots traditionally use a PID (proportional, integral, and derivative) control system, developed back in the 1920s. Some devices additionally adjust parameters in real time, based on the rate of course error or the rate of change in direction. This allows switching between "calm," "moderate," and "storm" modes, but remains essentially a mechanical process with no real learning curve.

Madintec MadBrain Controller Module

Because of this, many autopilots lose accuracy when navigating downwind or in strong gusts of wind, especially if the yacht is operated by a single crew. When artificial intelligence intervenes, the system begins to recognize patterns and select optimal control parameters. For example, the development Madintec MadBrain adapts the steering system's behavior to the specific yacht by analyzing how it reacts to waves, squalls, and changes in wind direction.

This powerful solution is actively used on IMOCA, Ultim, and Mini Globe 650 hydrofoil racing yachts. The technology is now being extended to high-speed cruising yachts and IRC offshore racing. The first four 33-foot Pogo RCs are already equipped with Madintec autopilots.

Faster than a yacht and a man

It's no surprise that artificial intelligence is actively used in the America's Cup, as it is in Formula 1. For example, QuantumBlack (the AI arm of consulting firm McKinsey) created an AI-powered virtual skipper for Emirates Team New Zealand ahead of the recent competition.

This system was trained by a professional Olympic sailor on a simulator, and the process took just one to two weeks. After training, the virtual skipper could evaluate up to 50 hydrofoil design options, taking into account various wind and sea conditions. What previously required months of on-water testing now takes just a few days. This significantly reduces the cost of developing America's Cup yachts.

Additionally, QuantumBlack has created physically based AI models that simulate the behavior of flows and other complex systems. This allows for the rapid testing of new ideas and design options, making the process much less expensive than traditional CFD (Computational Fluid Dynamics) calculations.

AI is also actively used in regatta broadcasts. For example, SailGP and the America's Cup use machine learning to recognize yachts, buoys, spectator boats, and the shoreline, creating 3D graphic overlays in real time. "When we know the helicopter's exact position in space, we can render 3D graphics directly on the video," explains Joseph Ozanne, head of simulation and AI at Alinghi Red Bull Racing.

Smart help

AI-enabled mooring systems, such as Raymarine DockSense and Volvo Penta Assisted Docking, use machine learning to accurately recognize objects such as pontoons, boats, and mooring structures. This helps the system accurately interpret the surrounding environment and makes maneuvering in marinas safer.

Raymarine DockSense

AI also simplifies the analysis of navigation and meteorological data, transforming vast amounts of information into understandable and visual conclusions. And perhaps in just a few years, every yacht owner will be able to run highly accurate weather models directly on their laptop using readily available software.

conclusions

AI will inevitably come to yachts—and in the foreseeable future, without being tied to the internet or the cloud. Local artificial assistants will take their place on the bridges of bulk carriers and tankers, smart autopilots will continue to assist racers... We think you can continue the analogies yourself.

The main thing is that AI remains a partner, an assistant, a loyal companion, but not a replacement for humans. After all, artificial intelligence is unlikely to be able to relax on the Côte d'Azur, so cruising yachting will likely remain the preserve of humans. However, a skipper would never refuse real AI assistance.

New video on Interparus channel

The Côte d'Azur beckons—and we've fallen for its charm. In this new episode, we'll tell you about a trip along the Côte d'Azur, the local yachting scene, and why you should visit it at least once in your life.