Norway’s Laser Robots Target Salmon Lice in Fish Farms

How submerged laser nodes in Norwegian salmon pens detect and shoot sea lice — the vision loop, the safety veto, and the real-world limits.

Norwegian salmon farms use submerged laser nodes that find sea lice on passing fish with machine vision and fire a short optical pulse to kill the parasite without harming the salmon. The best-known system comes from Stingray Marine Solutions, an Oslo-based company whose delousing nodes have been hanging in Norwegian sea pens since the mid-2010s; similar optical approaches have been pursued by other suppliers since. Sea lice are small parasitic copepods — mainly *Lepeophtheirus salmonis* — that attach to a salmon's skin and feed on mucus, skin and blood. They are the costliest biological problem in Atlantic salmon farming, and the laser node is one answer among several to a problem that chemical treatments stopped solving reliably.

Table of Contents

How an underwater delousing node actually works

The hardware is a sealed, pressure-rated housing lowered into a net pen and held at a depth where salmon naturally swim past. Inside are illumination, one or more cameras, a computer running detection models, and a steerable low-power laser on a fast optical mount. Power and data run up an umbilical to a surface unit or barge, because the duty cycle and compute load are well beyond what a battery would carry. The processing loop is the interesting part for a robotics reader.

Frames come in continuously; a detector has to separate a lice-shaped object from the fish's own markings, scale patterns, wounds, bubbles and drifting debris, then decide whether the target sits on the fish rather than floating in front of it. A stereo or depth-aware setup gives the range needed to focus the pulse at the right standoff distance. Targeting then becomes a moving-target intercept problem in a fluid. The fish is translating and rolling, the water column carries particulates that scatter the beam, and the system has a fraction of a second before the target leaves the frame. The laser is aimed by mirrors rather than by moving the whole head, because mirror galvanometers settle far faster than any gimbal carrying optics mass.

Why the salmon is not harmed

The physical argument rests on two differences between parasite and host. A sea louse is small, dark-pigmented and sits exposed on the surface, so it absorbs the pulse energy in a tiny volume. Salmon scales are reflective and the fish's body is large, so the same pulse spreads its effect across far more tissue and is partly bounced away.

Vendors pair that with an eye-avoidance rule: the vision system is trained to recognise the fish's head and eye region and to withhold a shot when the target falls near it. A pulse that would otherwise be harmless to skin is not harmless to an eye, so suppression logic matters more than beam power here. The practical limit is that this is a statistical safety argument, not a guarantee on any single shot. Fish welfare in a pen is assessed by sampling and inspection, so a farm running optical delousing still needs the ordinary welfare monitoring regime rather than treating the node as self-certifying.

What problem it replaces

Salmon farms have historically controlled lice with medicinal baths and in-feed treatments, then increasingly with mechanical and thermal delousing — crowding fish, pumping them through a vessel, and flushing or warming the lice off. Resistance to several of the common chemical treatments is widespread in North Atlantic salmon farming, which is what pushed the industry toward physical methods in the first place. The mechanical alternatives work, but they are rough. Crowding and pumping stresses the fish, can strip scales and mucus, and has been linked to mortality events during handling.

Each treatment is also a discrete, labour-heavy operation involving a vessel alongside the pen. A laser node changes the shape of the intervention rather than its goal. It runs continuously at low intensity instead of acting as an occasional high-stress event, so the aim is to hold lice counts under the regulatory ceiling rather than to crash a count that has already climbed. That suits the regulatory structure in Norway, where farms must keep average adult female lice per fish below a set threshold and report counts to the authorities, with a stricter limit during the spring period when wild salmon smolts are migrating past the farms.

The engineering constraints a demo never shows

The honest summary is that these systems are a suppression tool. They lower the rate at which lice counts climb and can reduce how often a farm needs a handling treatment, but farms generally run them alongside other measures — cleaner fish, skirts and snorkel barriers that keep salmon below the lice-rich surface layer, and conventional treatments held in reserve.

  • **Water clarity.** Turbidity from algae blooms, plankton and feed waste both degrades detection and attenuates the beam. Performance is a function of the week's water conditions, not a fixed spec.
  • **Biofouling.** Any optical window in seawater grows a film within days. Cleaning or wiper mechanisms are a maintenance line item, not an afterthought.
  • **Fish behaviour.** The node only treats fish that swim through its field of view at usable range. Salmon distribute themselves by temperature, light and oxygen through the water column, so node depth and pen coverage determine how much of the biomass is ever presented to the camera.
  • **Coverage arithmetic.** One node cannot watch a whole commercial pen holding on the order of a hundred thousand or more fish. Farms deploy multiple nodes per pen, and coverage per node is the number that drives capital cost.
  • **Power and connectivity.** Continuous operation across a pen array means cabling, surface infrastructure and a data link capable of carrying imagery or at least telemetry back to shore.

What a robotics reader can take from the design

This is one of the clearer commercial examples of closed-loop perception plus directed-energy actuation operating unattended in an uncooperative environment, and several design choices generalise. First, the system narrows its own problem physically before asking the vision model to solve it. Placing the node where fish naturally pass, controlling the illumination, and fixing the geometry removes variance that would otherwise have to be learned. Second, the safety behaviour is a hard veto on top of the detector rather than a property the detector is trusted to have learned — the eye-region suppression blocks a shot regardless of detector confidence.

Third, the actuator was chosen for settling time, which is the binding constraint in any intercept loop with a sub-second window. The cautionary lesson is about the metric. A node can report large numbers of successful shots while the pen's measured lice count barely moves, because shots fired is a system-activity metric and lice per fish is the outcome. Anyone evaluating a system like this should ask for the counted outcome on the farm, measured the way the regulator measures it, over a full production cycle.

Frequently Asked Questions

Does the laser kill the sea louse outright or just dislodge it?

The pulse is absorbed by the louse's pigmented tissue and destroys it; the dead parasite then falls away. It is not a mechanical dislodging method.

Can the same approach treat other fish species?

The detection models and safety rules are built around Atlantic salmon anatomy and sea lice. Applying it elsewhere means retraining detection and re-validating the eye-avoidance behaviour for a different body plan.

Do farms still need cleaner fish if they install laser nodes?

Most run both. Cleaner fish such as lumpfish and wrasse graze lice continuously across the whole pen, while nodes treat only fish that enter their field of view, so the two cover different gaps.

Is there a risk lice develop resistance to this?

There is no biochemical pathway to resist a physical kill, which is the main argument for optical and mechanical methods over medicinal ones. Behavioural shifts in where lice attach are a more plausible long-term concern than resistance.


You Might Also Like