About two minutes · no cost · no commitment
CorpusMarket Robotics
Robots in manufacturing: what is automated and what is not
Industrial robots have welded and painted in high-volume plants for decades. Most of what a small or mid-size shop does in a day has never been automated, because the parts, the batches and the setups keep changing. That hand work is what robotics labs now want on video.
By Linus Näslund, Co-Founder
- No cost to apply
- Nothing shared before we talk
- Reply within 2 business days
Contents07
Where robots in manufacturing stand today
Manufacturing is the one industry where robots are old news. Robots in manufacturing have been at work since long before anyone talked about AI, and in some plants they do most of the handling. They are also concentrated in a narrow kind of work.
- Industrial robots in fixed cells weld, paint, seal and palletize. They sit behind guarding and repeat one programmed path on parts that arrive in the same position every time.
- Collaborative robots (cobots) work next to people without a cage. They load and unload machine tools, drive screws and stack boxes.
- Mobile robots carry bins and pallets between stations.
- Camera systems check for defects that can be seen on a surface.
All of these depend on the same condition: the task has been made predictable. Someone designed a fixture, wrote a program and tuned it, and that effort pays off when the same part runs for months. Automotive body shops and electronics lines meet that condition. A job shop running forty parts on Tuesday and a different forty on Thursday does not.
What is still done by hand
Walk through almost any plant and the manual work is easy to find. It sits wherever the part, the material or the batch refuses to stay the same.
- High-mix, low-volume work. Short runs do not justify a fixture and a program for each part.
- Final assembly with flexible parts. Cables, hoses, gaskets, clips and trim bend and tangle. A person routes a harness by feel; a programmed arm has no feel to go on.
- Kitting and line feeding. Picking the right mix of loose parts out of bins.
- Rework and repair. Every defect is different, so every fix is.
- Inspection by touch. A finger across an edge finds the burr a camera missed.
- Changeovers and setups. Swapping tooling, clamping, shimming, running the first part.
The same pattern holds in the neighboring trades. In food manufacturing, machines fill, seal and palletize packaged product, while trimming, portioning irregular items, decorating and line sanitation are largely manual. In a cabinet or woodworking shop, a CNC router cuts the panels, and people do the assembly, sanding, finishing and hardware.
Humanoid robots in manufacturing are often presented as the answer to all of this. So far they are in trials, and the tasks shown publicly are simple material moves. A general-purpose robot that can route a hose or clear a jam has to learn it from examples, and those examples barely exist. How robots learn explains the method, and what physical AI is gives the wider picture.
What labs want recorded on a shop floor
Labs ask for ordinary tasks at ordinary speed, done by the people who do them every shift. Nothing is staged. Typical requests include:
| Area | Tasks labs ask for |
|---|---|
| Assembly | Fastening, fitting, cable and hose routing, clips, connectors, sub-assemblies |
| Machining and fabrication | Loading and unloading, deburring, measuring, fixture setup, tool changes |
| Kitting and packing | Picking loose parts, bagging, boxing, labeling |
| Rework and inspection | Checking by hand, sorting, small repairs |
| Food manufacturing | Trimming, portioning, tray packing, line changeover |
| Carpentry and woodworking | Assembly, sanding, edge work, finishing, hardware fitting |
Variety counts for more than volume of one task. A shop that changes jobs every day produces footage a single-product line cannot. If part of your operation is storage and shipping, the page on warehouses covers that side.
What recording looks like on a shop floor
In the programme we introduce businesses to, run by our partner Amelia Robotics, a worker wears a phone on a headband or hardhat mount with the camera pointing down at the hands. The worker starts a recording in the programme's app, does the agreed task and stops it. Recordings are reviewed, and the business is paid for the hours that are approved.
The camera sees what the worker's eyes see, which is why this is called egocentric data collection. Both hands stay free. A few things are specific to a plant:
- It is voluntary for each worker. Operators who do not want to record do not.
- Stations are chosen in advance. A shop usually starts with a few benches or cells where recording is simple, and leaves the rest alone.
- Existing rules come first. Where phones, loose items or head-worn gear are restricted, such as near certain machines or in food production areas, that is raised on the setup call and those areas are left out or handled as the programme and the business agree.
- The business decides the split. How pay is shared between the worker and the company is set by the business together with the programme.
The steps from first contact to first recording are on how it works.
What has to be settled first
A camera pointed at a workbench records more than hands. It can catch a customer's part, a drawing taped to the machine, a traveler with a part number, a screen showing a program, or a process the shop considers its own know-how.
How that is handled differs from one business to the next. A shop making its own product decides for itself what it is comfortable showing. A contract manufacturer often works under confidentiality terms with each customer, and those terms can differ by customer and by job. In practice some work ends up in scope and some does not: a generic deburring or packing task may be fine where a customer's assembly is not. Coworkers at the next station, who have not agreed to anything, are part of the same discussion.
None of this is decided by a general rule. It is gone through case by case for each business before recording starts. Our page on recording at work describes how these questions usually come up.
Does this put operators out of work?
People on a shop floor have watched automation arrive for a long time, so the question comes up quickly. The honest picture is that the tasks robots do well are the ones that were already engineered to be repetitive, and the tasks in the table above have resisted automation for decades for good reasons. Recordings help research on those tasks. They do not make a robot that can run a changeover next year. Many manufacturers also describe their problem as too few skilled people, not too many. We go through what is known and what is guesswork in will robots take jobs.
Next step for a plant or shop
If your people do hands-on work at a bench, a machine or a line and you want to know whether it fits, tell us about your business. It takes two minutes and commits you to nothing. We reply within 2 business days, and nothing is shared outside CorpusMarket before we have spoken with you.
Questions and answers
Frequently asked questions
What are robots used for in manufacturing?
Mostly repetitive, high-volume jobs in a fixed cell, such as welding, painting, palletizing, moving parts between machines and loading machine tools. The robot repeats a programmed motion on parts that arrive in a known position.
Are humanoid robots used in manufacturing?
They are at the trial stage. The pilots that have been described publicly are mostly simple material moves, such as carrying a tote or placing a part in a fixture. Skilled assembly, rework and setup are still done by people.
Which manufacturing tasks do robotics labs want recorded?
Hands-on tasks with a clear start and finish, such as assembly, kitting, cable and hose routing, deburring, packing, changeovers and inspection. The exact list is agreed with the programme before anyone starts.
Can we record if we make parts for other companies?
That depends on the job. Customer parts, drawings and processes are often covered by confidentiality terms, and some work can be recorded while other work cannot. It is gone through for each business before recording starts.
Do small shops qualify, or only large plants?
Small and mid-size shops are a good fit, because their work is varied and mostly manual. Whether a specific shop fits is checked when you get in touch.
Who gets paid, the shop or the worker?
The programme pays the business for recording hours that pass its quality review. How that is shared with the workers who record is set by the business together with the programme.
More pages
Read next
Warehouse robots: what they do and what they still need to learn
Warehousing is the most automated kind of work we cover. In the largest fulfillment centers, robots move shelves, sort parcels and pick items around the clock. The work they still cannot do is the long tail of awkward, varied tasks, and that is what labs want recorded.
Read moreHow robots learn: imitation learning and diffusion policy explained
Most robots that do something new today learned it by imitation learning, which means copying recorded examples of the task. This guide explains that method and the others around it without the math, and ends with what each one needs in data.
Read moreGet paid to record your work for AI
Robotics labs pay for video of people doing hands-on work, filmed from the worker's own point of view. This page is for businesses whose crews do that work every day. It explains what gets recorded, what the pay depends on and what to settle before anyone puts on a camera.
Read moreNext step
Tell us about your business
If your crew does hands-on work, a robotics lab may pay for video of it. Send your details and we will tell you whether it fits and what happens next.
- No cost to apply
- Nothing shared before we talk
- Reply within 2 business days