Agriculture and livestock
Harvesting robots and the difference between a pick and usable produce
Follow fruit selection, access, detachment and collection, then calculate marketable yield from a complete harvesting trial.
A harvesting robot must identify suitable produce, reach it, remove it and transfer it without unacceptable damage. The physical task depends on the crop and the way it is grown. Success on exposed strawberries does not establish success on fruit hidden behind leaves or on another crop.
The most useful report follows produce all the way to its accepted collection state. An image of a gripper holding fruit captures only one point in that process.
Follow the crop through the harvesting sequence
Harvest CROO describes a strawberry system with LiDAR-based row navigation, vision for evaluating berries and independently working harvesting units. Its technology page identifies ripeness and health assessment as part of selection. These are manufacturer descriptions; the page does not provide a complete independent success-rate dataset. [1]
The engineering sequence below helps identify what an evaluation would need to observe. The exact tool and movement must be checked for the selected harvester and crop.
- 1Find and classify
Locate candidate fruit and apply the stated ripeness and quality rule.
- 2Reach
Determine whether the tool can approach without unacceptable contact with the plant or nearby produce.
- 3Detach
Separate the fruit using the crop-specific method and retain control of it.
- 4Transfer
Move the fruit into the collection system without dropping or crushing it.
- 5Inspect
Count accepted produce and record misses, damage and rejected material.
A visible fruit and a reachable fruit are different cases
In a test, separate fruit that was never seen from fruit that was detected but not attempted. The first group can indicate a sensing or viewpoint problem. The second may expose access limits or a decision to avoid plant damage.
Record row geometry, cultivar, canopy condition and the way fruit is presented. A trained crop arrangement can be part of a practical system, but the report should identify the preparation rather than attributing every result to the robot alone.
A useful comparison uses the same acceptance rule for manual and robotic picking. Produce that would be rejected for a person should not count as a successful robotic harvest.
Calculate output from all eligible fruit
Consider an illustrative test with 200 eligible ripe fruits. The robot detects 180, attempts 160, detaches 144 and delivers 130 that meet the agreed quality rule. Detection is 90% of eligible fruit. Detachment succeeds in 90% of attempted picks. Marketable output is 130/200, or 65% of eligible fruit.
The two 90% figures could sound reassuring while leaving 35% of the intended harvest outside the accepted output. Keep the full sequence visible. Count unattempted fruit and damage separately so the next experiment can target the right stage.
These figures are invented for the calculation and are not results for Harvest CROO or another company.
Scroll sideways for all columns.
| Stage | Count | Interpretation |
|---|---|---|
| Eligible fruit | 200 | The denominator set before the run |
| Detected | 180 | Twenty eligible fruits were not detected |
| Attempted | 160 | Twenty detected fruits were not attempted |
| Detached | 144 | Sixteen attempted picks did not detach successfully |
| Accepted after transfer | 130 | The final accepted output |
Include field movement and collection in the time
Report elapsed working time alongside accepted quantity. A short picking cycle measured on one exposed fruit excludes travel between plants, search, retries, collection handling and stoppages. Parallel picking units also mean one unit's cycle time cannot be used directly as the whole machine's rate.
For a service proposal, request marketable output per elapsed hour under the proposed crop conditions. Record the support crew and any manual follow-up picking. This makes the remaining work visible instead of treating it as a free input.
A one-day result needs a seasonal boundary
A trial should state the date and crop condition. Ripeness distribution and access can vary across the harvesting period, so a single visit cannot establish output for an entire season.
Preserve original counts, quality rules and the timing of damage inspection. The grower can then compare results across visits and judge whether the operating window and collection logistics fit the farm. Agronomic decisions remain with the responsible specialists.
Sources and scope
Sources were consulted for this revision. Manufacturer descriptions are identified in the text. Worked examples are illustrative calculations, not measurements from a TechniaHQRobot test.
- Harvest CROO harvesting technology
Manufacturer description of strawberry navigation, vision and multiple harvesting units. No independent output rate is supplied here.