This is the design as it stands. It is a design, not a photograph — the model on this page is drawn from the specification, and no part of it has been built. Where a choice is still open, it says so.
What the building allows
A Dutch high-wire house is a fixed set of dimensions, and the machine is drawn around them rather than the other way round. Rows sit 1.60 m apart. The heating pipes double as the rail every trolley in the house already runs on. The aisle is shared with people, carts and hoses, and the crop moves as it grows.
- Two arms, forced by the pitch
- At 1.60 m between rows a single centre-mounted arm cannot reach both sides at a workable standoff. Two arms mounted either side of the centreline restore it. This is a consequence of the building, not an ambition — and two arms do not double the rate.
- The rail is table stakes
- Running the existing pipe rail collapses navigation to one degree of freedom and gives position for free. It is also what every trolley and every competitor already does. It is in the design because it is correct, not because it is novel.
- Battery as ballast
- The battery sits low and central, where the mass is useful. Swapping rather than charging is the intent, so the machine is not out of the row for hours. No swap dock exists yet.
One truss, start to finish
The sequence the machine repeats. Each step is a design intent; none of it has a timing figure attached, and it will not until hardware runs.
01
Look
Deck cameras sweep the row from the aisle and list the trusses worth approaching.
02
Approach
The arm moves the wrist camera to a standoff where the peduncle is resolvable through the leaf.
03
Verify
The wrist view decides ripeness and finds the cut point, or refuses the truss and moves on. Refusing is the cheaper error.
04
Cradle
The tool closes under the truss, taking its weight before anything is cut.
05
Cut
The blade severs the peduncle above the supported truss.
06
Place
The arm carries the truss to a deck crate and releases it.
07
Advance
The trolley moves along the rail. At the row end it changes aisle and starts again.
Two views, because one is not enough
A deck camera can find fruit but cannot be trusted to cut against — from the aisle the peduncle is behind leaf as often as not. A wrist camera can see the cut point but only once the arm is already committed to a truss. The design uses the deck view to decide where to go and the wrist view to decide whether to cut, and it is allowed to change its mind at the second step.
Ripeness read from colour depends heavily on which camera is asking and from how far. Any claim about how well this works has to name the camera, so until the new simulation runs there is no such claim on this site.
The map it builds on the way
The machine passes every plant in the house several times a week with cameras already pointed at the crop. Recording what they see costs nothing extra and produces a per-row record over the season: where fruit is setting, which rows are behind, and where a problem starts before it is visible from the path. The first version is plain colour imagery; thermal and multispectral are later.
Open questions
Written down because a grower or an engineer will ask, and because the list moving is how you tell the project is alive.
- Speed. Nobody in this category has a comfortable answer, and neither do we. It will not be answered on a slide.
- Two arms working the same aisle at once without either becoming an obstacle to the other. This is a geometry problem, not a software one.
- What a truss actually weighs, and what it takes to break a peduncle. Both are currently assumptions and both need a real plant and a scale.
- Crate handling and headland transfer, neither of which exists in any form yet.
- Whether the first module a grower will pay for is harvesting at all. One of the two growers we interviewed would rather have de-leafing.