TPFlabsTesting a harvesting robot on your own rows: what a pilot must prove
A robot demonstration tells you what the machine can do on the vendor's rows, on the vendor's fruit, on a good day. A pilot tells you what it does on your rows, in your season, with your pickers standing next to it. The difference is the only thing that decides whether the machine earns a place in your harvest plan, and it is worth structuring properly before anyone signs an order.
This is the structure we use in our own pilot programme, written so that it works for any selective harvesting machine, not only for ours.
Start with a question, not with a machine
A pilot without a written question turns into a long demo. Before the machine arrives, put the question on one page: what number decides the outcome? For most fresh-market blueberry farms it is one of these three.
Cost per kilogram picked, all in, including the idle months and the fruit the machine leaves behind. Kilograms per hour at your spacing, not at a demo spacing. Or the share of the crop the machine can touch at all without bruising: the fresh-market pack-out.
Choose one as primary and write the threshold that makes the pilot a success. Everything else you measure becomes supporting evidence.
Site requirements worth checking before the visit
Most pilot failures are site failures, discovered too late. Check these on your own field, with a tape measure, before the machine is booked.
Row spacing and crown shape. A machine built for 3.0 m rows will not enter 2.4 m plantings, and a planter with deep furrows can put the crown below the wheel line.
Ground and slope. Soft ground after rain, and any slope above a few degrees, change the traction and the reach of the arm.
Access at the row ends. Turning space is usually the real constraint, not the row itself.
Power and charging. If the machine runs a shift on battery, where the charger goes matters more than the battery specification.
Fruit presentation. Fruit hanging inside the canopy, or a heavy leaf load, changes the picking rate far more than the machine datasheet suggests.
The 30 / 60 / 90 structure
Thirty days: the machine picks a defined sample, supervised, on a small block. The goal is not output, it is the truth about the pick rate, the bruise rate and the failures. Count the interventions: every time a person stops the machine to free fruit, unjam the gripper or restart it, that is a data point, not an accident.
Sixty days: the machine works a normal shift with your crew, and the crew is told to treat it as one more worker who happens to be slow. The interesting number here is not the machine's rate but the change in the crew's rate, because the two share row ends, bins and tractors.
Ninety days: the machine runs the shift plan, and you compare cost per kilogram against your hand-picking baseline for the same block. If the pilot survives to this point, the decision is arithmetic, not opinion.
The data to collect
Keep the list short enough that someone actually fills it in.
Kilograms picked per machine hour, separated into clean fruit and culls, with the cull reason recorded. Bruise index, measured on a fixed sample per day by the same person. Downtime minutes and the cause, in your own words. Labour hours spent around the machine, including the operator and the person clearing jams. Fruit left on the bush at the end of the pass, estimated on marked plants. Energy per shift, if the machine is electric.
Five columns, one sheet, one person responsible.
Failure modes that end a pilot early
Bruise discovered on the packhouse line rather than in the field, which is why the bruise sample must be measured daily rather than at the end.
A pilot that runs only in perfect weather and never sees dew, dust or a wet canopy, which tells you nothing about your actual season.
No baseline: if hand-picking cost per kilogram is not measured on the same block in the same week, the comparison at day ninety is a guess.
Machine improvements arriving mid-pilot. A pilot measures one configuration; if the vendor ships a new gripper halfway through, the ninety days restart.
What a good vendor should give you
A written pilot plan with the metrics above, a named engineer who answers during the season, a spares kit for the wear parts, and the honest limits of the machine in writing before the first day. A vendor who will not put limits in writing has not tested the machine in a real season yet.
Our own pilot programme follows this structure, with the 30, 60 and 90 day phases and the same five metrics. If you want to see whether your block qualifies, the requirements and the current availability are on our pilot page.
Sources
See the sources section below for the public data behind the sector figures used here. Every cost number in this article is reproducible from them, and any figure we could not trace we left out.
Sources and data behind the figures
- ILVO — Institute for Agricultural, Fisheries and Food Research
- AgriFoodTEF — EU test and experimentation facilities for agri-food
- Eurostat — agricultural labour input and income indicators
- FAO — agricultural labour and harvest mechanisation
Cost models in this article are TPF Labs calculations. Sector inputs come from these public sources; each cost figure is reproducible from them.
What the machine costs and how a pilot works: pricing · pilot program