Preparing the published article.
Field Trials for Agricultural Robots: Plots, Controls, Records
How to design plots, blind control strips, and the exact field records that make a farm robot trial defensible at season end.
Define the Question and the Success Criterion Before the Machine Arrives
A field trial answers a question, so write the question before the machine reaches the gate. State the operational problem the purchase is meant to solve: fewer labour hours at a peak operation, earlier completion of a time-critical pass, lower input use on a defined block, or work that current practice cannot complete at all. Then write the success criterion in the same file, with a threshold, a measurement period, and the person responsible for measuring it. If the criterion is a feeling rather than a number, the trial will produce an argument rather than a decision.
- Which crop, variety, and growth stage will the machine work on, and which of them will it not be tested on?
- Which operation is being compared, done how, by which crew, at which pass?
- What is the minimum result that would justify a second trial?
- Which measurements are taken by the farm, and which are accepted from the supplier's log files?
- What happens if the machine cannot operate on a planned day, and who records the lost window?
Keep the criterion narrower than the ambition. A machine that reduces hand-weeding hours on one block over one season is a testable claim; a machine that improves soil health is not testable inside a season. Two useful distinctions protect the trial later: a demonstration proves that the machine can work, and a trial measures how it behaves in your conditions, including weather delays, breakdowns, and operator learning. Write down which one the supplier is offering, because the two have different evidence standards.
Lay Out Plots and Control Strips That Survive Tractor Traffic
Design the layout so it can be verified with a handheld device and a marked post, because verification that depends on memory fails by the second pass. Use paired strips: machine pass and control pass side by side across the same soil, the same drainage, and the same headland, each wide enough for one full harvester pass. Mark both ends of every strip with coordinates and with a physical marker, and record the marker type and the coordinate reference system used. One replication per field is not a design; draw the strips across the field's visible soil variation and repeat them in at least two fields before drawing any conclusion.
- Strip coordinates and a dated photograph pair, taken from the same standing point.
- Plot area, working width covered, travel direction, and turn pattern at the headlands.
- Machine settings per pass: speed, working depth, application rate, and who set them.
- Weather at application: rainfall before and after, wind speed and direction, soil condition by hand assessment.
- Compaction and traffic lines, including routes used by the tender or refill vehicle.
The most common flaw is an unequal control: a different variety, a different planting date, or an extra pass on the control explains a result better than the machine does. The second is placement by convenience, with the machine on the best soil and the control on the worst. Keep one distinction for the report: a treatment difference is not the same as a machine effect, so present them separately.
Record the Same Fields in Every Pass and at Harvest
Decide the harvest protocol before the season, because combining strips differently at harvest destroys the comparison. Harvest each strip separately and record fresh weight or yield per strip from the same scale or yield monitor, then convert to a per-hectare figure using the measured strip area. Note the moisture at harvest for each strip, since a wetter strip weighs more without yielding more. Ask for the yield monitor's calibration record and for the raw file, not only a summary table.
| What to record | How often | Why it changes the conclusion |
|---|---|---|
| Machine working hours | Every operating day | Converts a result into cost per hectare |
| Interventions and breakdowns | Every occurrence | Distinguishes machine reliability from crop response |
| Refills and payload per fill | Every refill | Shows the real limit on daily area |
| Operator identity | Every shift | Reveals a learning curve that flatters late-season results |
| Strip yield and moisture | At harvest, per strip | Provides the primary comparison |
| Crop condition and damage notes | Weekly per strip | Explains differences that yield alone hides |
Finally, agree who owns the raw data. A trial whose logs live only in a supplier portal cannot be re-examined, and a result nobody can re-examine is not evidence for a purchase decision.
Photo credits: the images in this article are licensed reference photographs of the relevant robot class, credited with author and licence in each caption; they are not photographs of the specific equipment discussed. Licence and author details are recorded with each image.