AI in underwater drones: what’s behind it, and how does it work?

5 October 2026

What this is about

“AI” now appears on almost every product page. For underwater drones it mostly means image recognition: a small computing model looks at the camera picture and recognises things in it, such as a diver, a fish or a weld seam. On top of that come smart sensors that help the drone stay in one place. We sort it into five areas.

1. Understanding images: recognising and tracking objects

How it works: A trained model examines the camera picture many times per second. It has seen thousands of example images beforehand and has learned what a diver, for example, looks like. When it spots one, it passes the position in the frame to the controller, and the thrusters correct so that the object stays in the centre.

• QYSEA AI Vision Lock: You tap an object in the app and the drone keeps it in frame. Motion sensors (acceleration and rotation) account for the drone’s own movement and cancel out drift.
• QYSEA AI Diver Tracking: Recognises divers and follows one from the side, behind, above or below, or keeps a whole group in frame.
• Blueye X7 and X3 Ultra: A built-in mini computer (Nvidia Jetson Orin NX) recognises objects directly on board, tracks a single object and helps to avoid collisions.
• CanFish CF1 (CHASING): The app alerts you as soon as fish appear in the picture, which is handy for fishing.

2. Improving images: haze and colour cast

Underwater, suspended particles and the water itself swallow light and colour. Filters can partly calculate this out: the plankton filter of the QYSEA E-GO and the haze filter of the new Blueye models remove distracting particles, and the simple TrueColor correction of the CHASING Dory brings back the reds. You can read more in our practical guide Colours underwater: why everything turns blue, and what helps.

3. Holding position: sensors instead of thumbs

Strictly speaking, this is less AI than clever sensing, but it feels just as “intelligent”. The simplest level is depth hold via the pressure sensor, which almost every drone can do. The professional level is called DVL (Doppler Velocity Log): a sonar measures how fast the drone is moving over the bottom. From that it calculates the direction and strength of the current and pushes back with its thrusters.

• CHASING X with “AnchorX”: holds against currents of up to 4.5 knots, according to the manufacturer.
• QYSEA E-MASTER: hovers via DVL in currents of more than 3 knots, with obstacle detection and return at the press of a button.
• QYSEA X1: uses two DVLs at once. You can read more in our news piece on the X1.
• CHASING Distance Lock Sonar (accessory): measures the distance ahead, to the side and downwards, and for example keeps a fixed distance from nets or pillars.

4. Building maps: sonar SLAM

SLAM stands for “simultaneous localisation and mapping”. A 3D sonar scans the surroundings, a computer builds a point cloud from it (in other words, a 3D map) and always knows where the drone is within it. This works even with zero visibility. It is newly in use on the Deep Trekker SPECTRA.

5. Measuring: lengths and areas from the picture

QYSEA’s QY-MT system analyses the 4K picture and measures lengths, angles, areas and curvatures underwater. With its stereo camera, the SPECTRA manages similar measurements without reference markers. For the hobby it is still a thing of the future, but for inspections it is already routine.

Who uses what? The overview

Conclusion

For hobby pilots, the most important AI feature today is object tracking in the style of Vision Lock: it makes steady footage much easier. Genuine current stability via DVL and sonar maps are so far found only on professional equipment, but experience shows that this kind of technology tends to find its way into cheaper models after a few years.

Lost something important underwater that a hobby drone can no longer recover? UDRA offers professional recovery missions in Austria.

EditorKennyKenny writes about ROVs, tests and missions for Unterwasserpilot.