Industries / Food, farming and forestry / Food and beverage processing / Seafood processing
Filleting, peeling and shucking
This is the work of cutting fish into fillets and portions and taking shellfish out of their shells: gutting, filleting, removing bones, peeling shrimp and opening oysters, scallops and lobsters.
7
Milestones achieved, from 1903 to 2018.
3
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1903
A machine guts salmon
Edmund A. Smith of Seattle built a machine that carried each salmon past knives and brushes that cut off the fins, split the belly and cleaned it out. Butchering was the last hand step left in salmon canning, and the machine replaced crews of Chinese butchers; its maker sold it under a racist name that the Alaska State Library records was used well into the 1900s. Sources disagree on its date and speed: the Alaska State Library gives 1902, first use in 1904 and 40 salmon a minute, and HistoryLink gives 1903 and 110 a minute.
Lessonthe machine had to be adjusted for different sizes of fish, a problem every later machine met.
Credit
Edmund A. Smith; Smith Cannery Machines Co
1921
Filleting machines from Lübeck
Rudolph Baader's firm in Lübeck, Germany, carried fish through its machines clamped by the tail and sold its first herring filleting machines in 1921. Cod was harder because sizes vary so much; the BAADER 99, made from 1951 after trials on British trawlers, sensed each fish's size with cam disks and adjusted its cuts for fish 50 to 120 cm long.
Lessonthe hard part was holding and measuring the fish, not the cut.
Credit
Rudolph Baader, Nordischer Maschinenbau Rud. Baader
1951
Shrimp peeled between rollers
The Lapeyre family of Houma, Louisiana, patented a machine that presses shrimp into the gap between powered rollers, which pull the shell away. A 2015 academic review names this roller type, and a clamp-and-cut type first patented by Gregor Jonsson in 1957, as the two leading designs, with one roller machine peeling about 400 to 450 kg an hour.
Lessonthe roller type cannot take out the vein, and the clamp type still needs an operator to lay out the shrimp.
Credit
Fernand, James M. and Emile Lapeyre (Peelers, Inc., now Laitram); Gregor Jonsson
2003
Water pressure opens oysters
Ernest Voisin of Motivatit Seafoods in Houma, Louisiana, patented a process, first filed in January 1998, that holds oysters in water at very high pressure to kill Vibrio bacteria. The pressure also releases the muscle that holds the shell shut, so the oyster opens without a knife. By 2003 the plant ran about 20 percent of its oysters through it, and a 2005 university test found 10 percent or fewer still closed after treatment.
Lessonthe pressure kills the oyster, so live oysters for the half shell are still opened by hand.
Credit
Ernest A. Voisin, Motivatit Seafoods; test by George J. Flick Jr
2006
Pressure frees raw lobster meat
John Hathaway's Shucks Maine Lobster in Richmond, Maine, puts live lobsters in a high-pressure vessel, which kills them within seconds and loosens the raw meat from the shell. Workers still crack the shells by hand, but the meat comes out whole and uncooked. The director of the Maine Lobster Dealers Association called the plant "the first of its kind" when it opened in 2006.
Lessonpressure solved the hardest step, getting the meat to let go, but not the cracking.
Credit
John Hathaway, Shucks Maine Lobster
2015
Water jets cut pin bones
Pin bones are the small bones along the middle of a whitefish fillet; in cod they are cut out by hand in a V-shaped strip, losing about 8% of the fillet. In the Nordic APRICOT project, Marel and SINTEF built FleXicut, which finds the bones by X-ray and cuts along them with a water jet. SINTEF's tests in February 2015 found it removed bones as thin as 0.1 mm with an average loss within the 6% target, and the first machines ran in Icelandic plants from January 2015.
Lessonmisses still need trimming and checks by people.
Credit
Marel; SINTEF; Norway Seafoods; Faroe Origin
2018
A machine shucks raw scallops
Nikko Co. of Kushiro, Hokkaido, built the Auto Sheller, which separates the shell, mantle and guts of raw scallops and keeps only the round white muscle that people eat, without cooking. It won the small and venture business prize at Japan's 8th Robot Award in 2018, cited for automating a product that varies from one scallop to the next. The company says one machine handles 96 scallops a minute, the work of about 11 people.
Credit
Nikko Co., Ltd
Watch for
Robots slicing raw salmon
Sashimi-Bot uses three robot arms to straighten a raw salmon loin, slice it with a knife held by one arm while another steadies the loin, and lift the slices onto a tray. Its shaping control was trained in simulation first; in tests on real loins it picked up 26 of 28 slices from the board and all 6 from the knife.
Achieved whenthe method trims or portions fish on a commercial line at production speed, or another lab reproduces it.
Claimed by
SINTEF Ocean, NTNU and partners, in the journal npj Robotics, June 2026
Shrimp peeled to hand quality
Laitram's general manager says the company has a technology roadmap to automate shrimp de-heading, peeling, deveining and camera grading "to the quality of hand-peeled shrimp", and is beginning to sell it, particularly in Ecuador and India.
Achieved whena processor or an outside party in Ecuador or India publishes yield and quality figures for the machines against hand peeling.
Claimed by
James Lapeyre, Laitram, September 2025
Robots cutting out worms
Nofima tested ways to remove worms once a camera has found them. Water-jet cutting works at standard line speed, about 40 cm a second, but cuts away a whole region; a robot that removes single worms worked only as a proof of concept, with "low reliability, tracking errors" and too little speed.
Achieved whena robot removes single worms on a production line at line speed and a party outside the developers reports the yield.
Claimed by
Nofima, March 2026
Open
01
Mapping the oyster's muscle
A public set of oyster 3D scans or photos with the hinge and the adductor muscle, the muscle that holds the shell shut, marked after hand shucking, and a model that predicts from the outside of the shell where that muscle sits. A machine that opens live oysters needs this map first.
The briefCredit · unclaimed02
Agreeing on what to trim
A public set of raw salmon or whitefish fillet photos on which several trained trimmers each mark blood spots, dark pigment, leftover membrane and fin, and how much they would cut; a measure of how far they disagree; and an image model that makes one consistent call.
The briefCredit · unclaimed
Open 01 · the brief
Mapping the oyster's muscle
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of oyster 3D scans or photos with the hinge and the adductor muscle, the muscle that holds the shell shut, marked after hand shucking, and a model that predicts from the outside of the shell where that muscle sits. A machine that opens live oysters needs this map first.
Why it is still open
Oysters for the half shell are opened alive, one at a time, by skilled workers with knives; pressure shucking kills them. A University of Maryland student team building a shucker says shucking "requires skilled, time intensive labor" and reported "occasional inaccuracies in adductor muscle locations". Searches found research prototypes from 2007 onward but no public map of shell shape to muscle position.
Built with what exists
A phone camera or a structured-light 3D scanner such as a Revopoint POP 3, an oyster knife and a cut-resistant glove for the reference shucking, Meta's open Segment Anything 2 model to outline shells, and a small model trained in PyTorch.
First step
Buy two dozen oysters, photograph each from above and from the side on a ruler, shuck them, and photograph the muscle scar inside each shell to see how far it moves from one oyster to the next.
Done looks like
500 oysters of at least two species scanned, shucked and labelled, with the data published; on 100 oysters held back from training, the predicted muscle centre falls within 5 mm of the real one for at least 90%.
Open 02 · the brief
Agreeing on what to trim
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of raw salmon or whitefish fillet photos on which several trained trimmers each mark blood spots, dark pigment, leftover membrane and fin, and how much they would cut; a measure of how far they disagree; and an image model that makes one consistent call.
Why it is still open
A 2022 review by Marel staff says trimming fillets and quality inspection still "rely on manual labor", that "the perceived severity of a defect can vary highly between operators", and that trimmers tend to over-trim to avoid complaints, which costs yield. Searches found defect-detection research on salmon fillets from 2007 and 2016 but no published data set with several trimmers' calls on the same fillets.
Built with what exists
A colour camera over a bench with even lighting, Label Studio for marking, Meta's Segment Anything 2 fine-tuned on one graphics chip, and a kitchen scale to weigh the trim.
First step
Buy 20 salmon fillets with visible blemishes, photograph them, and ask three fish-counter workers to mark on printouts what they would cut.
Done looks like
300 fillets, each marked by at least three trimmers; their agreement reported for each kind of defect; the model's marks agree with the trimmers' majority at least as often as one trimmer agrees with another; data published.