Use case · Operations & production

Vision-guided automated processing

2D and 3D vision measure each variable natural item, such as a carcass, fillet, log or piece of produce, and AI picks the cut path or process action that automated equipment then carries out.

Maturity
Proven, scaling
Complexity
High (our assessment)
Department
Operations & production
Technology
Computer vision, Robotics & autonomy, Machine learning
Solves
Hard-to-fill roles, Scrap & rework, Safety incidents, Throughput bottlenecks

Evidence at a glance

2documented deployments
2manufacturers
0independently assessed
Independently assessedManufacturer-reportedVendor-reported

Who has done it

Cargill · Dodge City beef plant

A 3-D vision system scans meat in a cradle and finds the best cutting path for an automated bandsaw that separates meat from bone.

Manufacturer-reported
Dodge City, KS · 2024
Making three thousand fewer cuts by hand each dayeach automated saw at our Dodge City plant can make up to 14 cuts every minute. This leads to higher output in less time

Tyson Foods

Computer vision and machine learning are being tested to guide robotic meat cutting.

Manufacturer-reported
· 2018
View the data as a table
ManufacturerEvidenceOutcome as reported
CargillManufacturer-reportedMaking three thousand fewer cuts by hand each day
CargillManufacturer-reportedeach automated saw at our Dodge City plant can make up to 14 cuts every minute. This leads to higher output in less time

What it is

Vision-guided automated processing pairs cameras, 3D scanners and sometimes X-ray with AI models that work out, for each individual item, how it should be cut, trimmed, sorted or handled, and then steer a robot, water-jet cutter, saw or blade to carry out that decision. It is used where the raw material is natural and no two pieces are alike: meat and poultry deboning and trimming, fish portioning, fruit and vegetable coring or grading, and log and lumber cutting. The system measures shape, size, fat and bone position or defects on every piece and computes a cut path that balances yield, product specification and cycle time.

It matters because these operations have long depended on skilled manual labor with knives and saws, which is hard to staff, physically demanding and a leading source of cuts and repetitive strain injuries. Fixed automation struggles because it assumes every piece is the same. Measuring each item and deciding its cut individually lets equipment adapt to natural variation the way an experienced worker does, and it records yield data for every piece.

How it works

  1. Choose the task and the value

    Engineering and operations pick a station where labor is hardest to fill, injuries are frequent or yield losses are large, and agree what a good cut means in terms of yield, trim specification, bone or defect tolerance and throughput.

  2. Scan every item

    Cameras, laser or structured-light 3D scanners and, where internal structure matters, X-ray capture each piece as it enters the cell. Sanitary enclosures, washdown ratings and lighting control are part of the design in food plants.

  3. Decide the action

    Models trained on scanned items and expert-labeled cuts locate landmarks such as joints, bones, fat seams or defects, then an optimization step chooses the cut path or portioning plan that best meets yield and specification for that specific piece.

  4. Steer the equipment

    The plan is sent as tool paths or setpoints to robots, water-jet or blade cutters, or sorting gates, synchronized with conveyor position so the cut lands where the scan said it should.

  5. Check and learn

    Downstream checks, weigh scales and operator inspection compare actual yield and trim quality to the plan. Misses are labeled and fed back into training as raw material, seasons and suppliers change.

Benefits

Less dependence on scarce skilled labor

Stations that needed experienced cutters on every shift can run with fewer people, who move to monitoring, quality checks and finishing work that still needs human judgment.

Higher, more consistent yield

Cut paths computed for each piece follow its actual shape and structure, which recovers more saleable product than cuts based on averages or on a tired worker's judgment late in a shift.

Fewer hand and strain injuries

Moving knife and saw work into guarded automated cells removes people from some of the highest-risk tasks in food and wood processing.

Piece-level production data

Every item's dimensions, cut plan and yield are recorded, which helps purchasing evaluate raw material suppliers and helps operations spot drift in equipment or incoming material.

Data & prerequisites

Data you will need

  • 2D images, 3D scans and, where used, X-ray images of each item entering the cell
  • Expert-labeled examples of landmarks and correct cuts across raw material variation
  • Product specifications, trim standards and value by cut or portion
  • Conveyor encoder and robot or cutter position data for synchronization
  • Downstream weights, yields and quality inspection results per item or batch
  • Raw material lot and supplier information

Before you start

  • A clear, written specification for each cut or portion that experienced workers agree on
  • Engineering capacity for sanitary or industrial cell design, guarding and controls integration
  • Stable product presentation on the conveyor so items can be scanned and tracked
  • A plan for retraining or redeploying workers and for maintaining the equipment

Common pitfalls

  • Underestimating raw material variation across seasons, suppliers and animal or crop sizes, so a cell tuned in testing underperforms later
  • Treating sanitation, washdown and food safety requirements as an afterthought in the cell design
  • Measuring success on throughput alone and missing yield losses that cost more than the labor saved
  • Not involving experienced cutters in labeling and acceptance, which produces cuts the plant does not consider correct

Questions to ask vendors

  1. What range of item sizes, shapes and conditions does your system handle, and what happens to items outside that range?
  2. How do you measure yield and trim quality during acceptance, and on what volume of production?
  3. Which sensing do you use, 2D, 3D or X-ray, and why is it needed for our product?
  4. How does the cell meet our sanitation, washdown and guarding requirements?
  5. How do we add a new cut, product or specification, and does that require your engineers?

What to measure

  • Yield per item or per unit of raw material
  • Labor hours per unit at the automated station
  • Recordable injuries at the processing step
  • Throughput per hour and cell uptime
  • Out-of-specification portions or trim defects found downstream

Frequently asked questions

Why can't conventional automation handle meat, produce or wood?

Conventional automation repeats the same motion on parts that are the same. Natural items vary in size, shape and internal structure, so a fixed cut either wastes product or misses the specification. Vision-guided systems measure each piece first and compute a different cut for each one.

Is this the same as visual inspection?

They share cameras and models, but inspection decides whether a part is good or bad, while vision-guided processing decides what action to take on each item and steers equipment to do it. Many cells also inspect, for example rejecting items with defects before they are cut.

Does this replace skilled workers entirely?

Usually not. Automated cells take over the most repetitive and hazardous cuts, while people handle exceptions, finishing, quality checks and equipment care. Experienced workers are also needed to define and label what a correct cut looks like.

Related use cases

Quality

Visual quality inspection

Cameras and computer vision models inspect parts on the line to catch surface, assembly and dimensional defects consistently.

Computer visionMachine learning
29 documented deployments8 software companies

Operations & production

AI robotic picking & manipulation

Vision-guided robots pick, place, sort and handle varied or unstructured items that fixed-program automation cannot.

Robotics & autonomyComputer visionMachine learning
4 documented deployments2 software companies

Operations & production

Process & yield optimization

Models recommend or automatically adjust process setpoints to keep quality on target while improving yield, throughput and energy use.

Machine learningOptimization & planningIndustrial IoT & edge
34 documented deployments7 software companies

Safety, health & sustainability

Worker safety & EHS vision analytics

Computer vision on existing cameras spots unsafe conditions such as PPE gaps, pedestrian and forklift conflicts and zone intrusions.

Computer visionMachine learningGenerative AI & LLMs
7 documented deployments4 software companies

Put AI to work

Find the AI opportunities hiding in your plant.

Genvo, our implementation partner, works with manufacturers to pick the right first project, connect the systems you already run and put AI agents and tools into daily use, with people approving what goes out.

Assessments and consultations are provided by Genvo. How our partnership works.