Industries / Food, farming and forestry / Food and beverage processing / Meatpacking
Cutting and deboning
Cutting and deboning is the work of breaking a chilled carcass into primals, the large sections such as the loin or shoulder, then into smaller cuts, and separating the meat from bone, fat and skin.
5
Milestones achieved, from 1853 to 1994.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1853
Hogs cut in seconds
The writer Frederick Law Olmsted watched the cutting table of a Cincinnati pork house on a trip through the city in November 1853. Two men lifted and turned each hog and two cut it with cleavers into hams, shoulders and other cuts, which helpers sent on by trucks and dumbwaiters. He timed 35 seconds from one hog landing on the table to the next and called the team "a sort of human chopping-machine".
Credit
Cincinnati pork packers; account by Frederick Law Olmsted in A Journey Through Texas (1857)
1967
Beef ships in boxes
Iowa Beef Packers (IBP) opened its first plant in Denison, Iowa, in 1961, with a line on which each worker repeated one task. In 1967 it opened a plant at Dakota City, Nebraska, to make boxed beef: instead of shipping whole carcasses, it cut away fat and bone at the plant and shipped vacuum-packed cuts to the retailer's specification. Wholesalers and supermarkets no longer needed their own butchers, and meatpacking moved from city stockyards to small towns near the feedlots of the Great Plains.
Lessonmoving the cutting into the plant took skilled butchering out of stores and replaced it with single repeated tasks on a line.
Credit
Iowa Beef Packers, later IBP Inc
1990
Robots try beef boning
From 1987 a UK research program funded by the Agricultural and Food Research Council worked on robots that separate meat from the ribcage of a beef forequarter, using force and vision sensing, and published results in 1990. Koorosh Khodabandehloo, who led the work, wrote in 2022 that robot meat cutting had "met with exceptional challenges over the past 4 decades" and that many projects were still at the research stage. In 2016 Dale Woerner of Colorado State University said there had not been many successes with beef because cattle vary too much in size, shape and composition, while pork is simpler to cut.
Lessonby Woerner's account, the difference between one animal and the next is what holds back beef robots.
Credit
Koorosh Khodabandehloo, then at the University of Bristol, with N. Maddock and G. Purnell
1990s
Machines debone chicken legs
After a Japanese food processor asked for a machine to replace deboning whole chicken legs by hand, Mayekawa built its first automatic leg deboner, which Poultry World dated to about 20 years before 2012. Its second model, the Toridas Mark 2, deboned up to 1,000 legs an hour, and more than 650 of the machines had been installed worldwide by 2012. Mayekawa says it delivered its 2,000th Toridas in 2019.
Credit
Mayekawa Mfg. Co
1994
Machines strip bones of meat
Advanced meat recovery (AMR) machines press the meat left on bones, such as split neck bones, away from the bone. In 1994 the US Food Safety and Inspection Service (FSIS) let their output be labeled "meat" if its calcium stayed below a limit, a sign that bone had not passed through. A 2002 FSIS survey found spinal cord or dorsal root ganglia, nerve tissue that can carry the agent of mad cow disease (bovine spongiform encephalopathy, BSE), in final beef AMR product at 25 of 34 plants tested and in 89 of 256 samples (35%). After a cow in Washington State tested positive for BSE on 23 December 2003, an emergency rule of 12 January 2004 barred the skulls and spines of cattle 30 months and older from these machines and kept such nerve tissue out of anything labeled meat.
Lessonthe calcium test measured bone, so it could not catch the nerve tissue that lies next to the bone.
Credit
FSIS
Watch for
Camera coaches hand boners
Cargill says its CarVe system, an AI camera system with a display at the boning workstation, shows workers in real time how much good meat is still on the bone, so that they can change their knife strokes on the spot. Leon Fletcher, Cargill's vice president of operations, said early trials showed yield gains of up to 0.5%.
Achieved whena party outside Cargill publishes yields from plants or shifts with and without the system, measured the same way.
Claimed by
Cargill, February 2026
Robot saw takes chine bones
The chine bone is the part of the spine left on a beef striploin. It is cut off by hand on a band saw, and a cut too deep loses meat. Frontmatec, which already sells a robotic chine bone saw for pork, trialed a beef version with AMPC at JBS Australia's Brooklyn plant. JBS said robot-cut loins, boned and trimmed as usual, gave striploin yields about 0.3 percentage points of hot carcass weight above its 12-month averages and were cut more evenly than by band saw; it called these early results from a short trial.
Achieved whenthe planned second-site trial is done and yields are compared with same-day manual cuts and reported by a party outside the companies.
Claimed by
Frontmatec, JBS Australia and AMPC, November 2026
Open
01
Record expert boning with forces
An open dataset of skilled boners deboning one standard cut, such as a pork shoulder or a lamb leg, with a knife whose handle measures force and twist, tracking of the knife's position and angle, wrist and overhead video, and the weighed meat, bone and trim from each piece. With it, a first robot model trained to predict the next knife stroke.
The briefCredit · unclaimed02
Test deboner yield claims publicly
A published side-by-side test of deboning chicken legs or thighs by machine and by a trained hand crew, on matched birds from the same flock and weight band, weighing meat, bone and trim and counting bone fragments left in the meat.
The briefCredit · unclaimed
Open 01 · the brief
Record expert boning with forces
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open dataset of skilled boners deboning one standard cut, such as a pork shoulder or a lamb leg, with a knife whose handle measures force and twist, tracking of the knife's position and angle, wrist and overhead video, and the weighed meat, bone and trim from each piece. With it, a first robot model trained to predict the next knife stroke.
Why it is still open
Robots that learn from demonstrations need recordings of experts, and searches of the web and GitHub for this file found none for deboning that pair knife motion with cutting force. The nearest is Origine AI's MEAT-CUT sample on Hugging Face: 214 video clips of 1.8 seconds each from a French butchery, with no force data, under a non-commercial, no-derivatives license, with the full 50-hour set sold commercially. A 2024 feasibility study by Aly and colleagues logged knife forces only for 20 mm cuts across striploin chops. Instrumented knives are not new: Dempsey and McGorry (2004) measured the grip forces and cutting moments of nine workers deboning pork shoulders, to judge the strain on their hands, but no released recording of such forces with knife position and video turned up.
Built with what exists
A six-axis force and torque sensor fitted in a knife handle (for example from ATI Industrial Automation or Bota Systems), ArUco markers and two or more cameras, or an OptiTrack system, for knife pose, an inertial sensor on the wrist, Hugging Face's LeRobot dataset format (see the Note "No download"), and a baseline model such as NVIDIA's GR00T 2 (see the Note "Five demonstrations per task").
First step
Fit a force sensor into a boning-knife handle, check it against weights, and record one boner deboning ten lamb legs at a butchery school or university meat lab.
Done looks like
At least 100 pieces deboned by at least three skilled boners, each with synchronized video, knife pose to within 2 mm of a reference measurement, force logged at 500 samples a second or more, and weighed yields, released under an open license; plus a baseline model scored on how closely it predicts the knife path of a boner it was not trained on.
Open 02 · the brief
Test deboner yield claims publicly
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A published side-by-side test of deboning chicken legs or thighs by machine and by a trained hand crew, on matched birds from the same flock and weight band, weighing meat, bone and trim and counting bone fragments left in the meat.
Why it is still open
The claims conflict and come from sellers. In a 2010 paper promoting their own YieldPlus deboning line, Bettcher Industries and Gainco wrote that automated thigh deboning yields were "typically stuck in the low-to-mid 60% range", with bone fragments and rework, and put hand lines at about 55% to 75% of input weight depending on skill. Mayekawa told Poultry World in 2012 that its machine's yield is "as good as that done by a skilled human". Searches of the web and GitHub for this file found no independent, published comparison.
Built with what exists
A plant's existing automatic deboner and hand line, calibrated platform scales, an X-ray inspection unit or trained hand checks for bone fragments, and R or Python for the statistics.
First step
Write the matched-sample protocol and ask a university poultry science department with a pilot plant, or a processor that runs both hand and machine deboning, to host a one-day pilot of 200 legs per method.
Done looks like
At least 2,000 legs per method over three production days, with meat yield as a percentage of leg weight and its 95% confidence interval, bone fragments per 10 kg of meat and labor minutes per 100 legs, published with the protocol and raw weights, and repeated at a second plant.