← sources

Building an optical sorter for potato harvest

source: reddit (reddit:robotics)published 2026-09-20 13:19ingested 2026-09-21 02:38

retained text

Hello all,

I’m a farmer from the Netherlands and I am interested in robotics. I am starting to work on a project that I would like to get feedback on.

During the potato harvest a lot of soil clods come in with the potatoes. At this moment, the only reliable way for me to remove them before the potatoes go into storage is with people standing alongside a conveyor belt picking them out. I would like to see if I can automate that process with a machine that I can build myself.

Since tools like ChatGPT and Claude became available, I have developed a habit of thinking: “How hard can it be? I’ll just build it myself.”. This approach has the danger of discovering halfway through a project that there is an entire layer of complexity that I didn’t even know existed.

So, before I start buying cameras, pneumatic components and machines, I would like a reality check from people who know more about robotics than I do.

**Current solutions**

There are commercial optical sorting machines that can already do what I need.

For example:

Flikweert Divider Select

[https://flikweertvision.com/nl/machines/divider-select/](https://flikweertvision.com/nl/machines/divider-select/)

 Downs Cropvision

[https://www.downs-fr.com/produits/downs-cropvision/](https://www.downs-fr.com/produits/downs-cropvision/)   

For my operation these machines are currently too expensive to financially justify. And I also think I only need a fraction of what these machines can do.

There is also a much simpler machine for separating clods from potatoes This type of machine uses a steel roller. Potatoes and clods rebound from the roller with different trajectories because of their different physical properties. An adjustable divider then separates the two streams.

Traditional clod separator

[https://www.youtube.com/watch?v=d\_j7X2TJx3M](https://www.youtube.com/watch?v=d_j7X2TJx3M)

This sounds like the solution to my problem, but I want to remove the clods directly during harvest, before the potatoes go into storage. At that point our clay clods can still be wet and soft. They do not behave sufficiently differently from a potato when they hit the steel roller, so the mechanical separation becomes much less reliable.

**The project**

The practical target would be to process 20–30 tons/hour over a 1-meter-wide belt. Potatoes and clods are generally up to around 70 mm. The system does not need to be perfect. Missing some clods is acceptable, and occasionally rejecting a potato together with a clod is also acceptable. I have divided the project into 3 parts: Mechanical hardware, vision/ejector hardware and software.

I’m most comfortable with the mechanical part of the project. The frame, conveyors, motors, mounting brackets, compressed air, etc., I can fabricate and modify this kind of equipment myself.

Instead of building the whole mechanical machine from scratch, I’m considering buying one of those traditional clod separators and using it as the mechanical base (see picture 1 and 2). The machine that I would like to use as a basis has an input conveyor belt that is 1-meter-wide and already has the two discharge paths I need.

https://preview.redd.it/arpbeh76doqh1.jpg?width=300&format=pjpg&auto=webp&s=ec5facf2e86f447b6d739ba14e991870e438fcda

Picture 1: Traditional clod separator

https://preview.redd.it/j8h5nj76doqh1.jpg?width=301&format=pjpg&auto=webp&s=cb1e522703c2308bb275ffd781d39036e62039be

Picture 2: Traditional clod separator

My idea would be to remove the steel roller and replace it with a row of individually controlled pneumatic ejector fingers (see picture 3). Above the 1-meter input conveyor would install a camera with controlled lighting (see picture 4).

https://preview.redd.it/kcc84h76doqh1.jpg?width=410&format=pjpg&auto=webp&s=d35e1872bcdcaaff61362c6c979f3757d2607cbc

Picture 3: Pneumatic fingers

https://preview.redd.it/tesjzf76doqh1.png?width=372&format=png&auto=webp&s=70e37da4a2e0a2941fcd255a93713410d3fa70da

Picture 4: Cameras and controlled lighting

The vision system needs to distinguish between a potato and a clod, determine its position on the belt, and eventually trigger the correct pneumatic finger when the clod reaches the end of the conveyor. Picture 5 shows roughly the view a camera would have of the conveyor belt with potatoes and clods.

https://preview.redd.it/ztlx5h76doqh1.png?width=605&format=png&auto=webp&s=f8bc97d1485c329b4df4bd5d029a1efe13d7caad

Picture 5: Conveyor belt with potatoes and clods

When it comes to vision and ejector hardware I am thinking of using 1 or 2 global-shutter cameras mounted on top of the 1-meter feed conveyor in an enclosed environment with LED lighting and 20 pneumatic ejector fingers across the width of the conveyor belt.

My questions would be: Are 1 or 2 global-shutter cameras a realistic starting point for this kind of application? What kind of cameras would you suggest? Would you build this from standard pneumatic cylinders and valves, or are there existing modules/components that would make much more sense for my situation?

When it comes to the software, I’m not a professional programmer so I would need to rely as much as possible on existing camera SDKs, vision frameworks and libraries rather than inventing everything from scratch.

Conceptually I imagine the system as: Camera → detect potato/clod → determine position → track conveyor movement → select ejector finger(s) → fire at exactly the right moment.  First a vision computer decides what needs to be rejected and where it is.  Second, an encoder tracks how far the conveyor has moved. Lastly a PLC/Real-time controller handles the timing of the pneumatic valves.

My questions would be: Is this realistically something I can build myself without professional programming experience, or am I underestimating how much expertise this requires? Does my concept make sense, or am I underestimating the integration and timing problem of the vision computer, encoder and PLC controller?

How hard can this be? I understand very well that this is not a simple project and that is the reason why I’m posting here first. I would like to get your feedback and insights on my ideas and project.

Thank you! 

 

document fetched 2026-09-21 00:33 · canonical https://reddit.com/r/robotics/comments/1wlgx3p/building_an_optical_sorter_for_potato_harvest