Agriculture & Future Food Systems · Article

Robotics and system failure

Evaluate an agricultural robot by the task it completes, the crop it affects and the conditions under which it can fail.

Preparation and related reading

Shared Foundations

This article helps you

Separate perception, movement and actuation.
Evaluate desirable task outcomes and crop damage separately.
Describe downtime and maintenance without equating availability with safe performance.

Define the task

A robot described as a weeder may identify plants, navigate between rows and operate a tool. These are linked tasks, but success at one does not prove success at the others. A useful assessment names the crop, weeds, growth stage, field conditions and the outcome being measured.

For example, detecting a weed in an image is different from reaching it accurately or controlling it after treatment. Removing a crop plant is an undesirable result even if the system removed many weeds. The comparison must preserve both outcomes rather than hiding crop injury inside a single attractive score.

Perception, movement and action

Perception estimates what and where something is from sensor information. Navigation determines how the machine moves; actuation produces a physical action, such as operating a mechanical tool. Timing and alignment connect these functions: a correctly recognized target can still receive a misplaced action.

Research systems illustrate the dependencies. USDA’s BenchBot gathers field images for analysis and model training. A primary narrow-row robot study combined cameras, navigation, targeting and autonomous recharge. Image collection is not itself weed control, and a bounded research demonstration is not proof of reliable operation in every production setting.

Conditions can break the chain

The narrow-row study described illumination-dependent image processing and mobility problems on muddy soil. Its field spraying tests used coloured liquid to observe targeting. Such tests can assess a delivery function without demonstrating weed mortality, chemical suitability or crop safety.

Changes in the scene, traction, power, positioning or tool condition can affect different parts of a system. Test the conditions relevant to the intended use, including what happens when information is missing or a component stops responding. Performance documented for one tested situation does not establish performance for another.

Separate removal and crop harm

Consider an invented inspection of the same 200 plants before and immediately after a mechanical pass: 100 weeds and 100 crop plants. Eighty weeds are removed and 20 remain. Five crop plants are removed and 95 remain. Assume the before-and-after classifications and counts are correct for this teaching case.

Weed removal is 80/100 = 80%; crop removal is 5/100 = 5%. The denominator is the corresponding initial plant group. The diagram and table preserve these separate accounts. This immediate count does not measure later regrowth, crop yield, hidden injury or overall ecological effects.

Invented immediate removal outcomes for two initial plant groups
Initial group	Removed	Remaining	Removal fraction
100 weeds	80	20	80% of weeds
100 crop plants	5	95	5% of crop plants

Two initial groups, separate outcomes

Arrows partition each initial count; removed plants do not turn into remaining plants. Invented counts, no tested robot performance claim.

Read the diagram as text

In an invented immediate account,100 weeds divide into 80 removed and 20 remaining;100 crop plants divide into 5 removed and 95 remaining. The groups retain their own denominators.

Parts

100 weeds initially: Invented before-pass count.
80 weeds removed: 80% of the initial weed group.
20 weeds remain: Immediate remainder; later regrowth unknown.
100 crop plants initially: Invented before-pass count.
5 crop plants removed: 5% of the initial crop group; an undesirable result.
95 crop plants remain: Immediate remainder; other injury not assessed.

Connections

100 weeds initially → 80 weeds removed: removed
100 weeds initially → 20 weeds remain: remain
100 crop plants initially → 5 crop plants removed: removed
100 crop plants initially → 95 crop plants remain: remain

Availability and work completed

In an invented four-hour assessment window, a robot is available for task operation for three hours and unavailable for one. Time availability is 3/4 = 75% under that definition. It does not report the area successfully treated, the accuracy of action, or safety during the available hours.

Record why operation stopped and how long detection, recovery, recharge and maintenance took. A system that needs human intervention still needs that support counted. Compare complete task performance, labour, energy, damage and upkeep on a common basis; elapsed availability alone cannot establish affordability or efficiency.

Failure response and maintenance

Automation changes responsibilities rather than removing them. The system needs a defined response to lost sensing, communication or control, and people need appropriate competence to assess faults. A stopped display or drive is not proof that stored energy and movement hazards have been removed.

Great Britain’s HSE guidance stresses competent maintenance and control of equipment hazards, including stored energy and unintended restart. This article supplies no isolation sequence or authorization for unattended operation. Equipment-specific instructions and applicable requirements govern real work; a research performance score cannot substitute for them.

Worked examples

Keep two plant groups visible

In the invented before-and-after account, what are the weed-removal and crop-removal fractions?

Weed removal = 80 removed weeds/100 initial weeds × 100 = 80%.
Crop removal = 5 removed crop plants/100 initial crop plants × 100 = 5%.
The remaining counts are 20 weeds and 95 crop plants; each group balances its initial 100.

The immediate results are 80% weed removal and 5% crop removal, with later control and production outcomes unknown.

Calculate a defined availability

A robot is available for task operation for three hours in a four-hour window. What does the percentage describe?

Use the defined observation window: 3/4 × 100 = 75%.
The remaining hour is 25% of that window.
Neither percentage measures safe operation, treated area or plant outcomes.

Time availability is 75% for this window and definition, not a complete performance result.

Check your understanding

A camera recognizes a weed, but the tool acts beside it. Which distinction explains the failure?
Show answer and reasoning
Recognition is separate from successful actuation.
Position, alignment and timing must connect perception to the physical task.
A field test shows coloured liquid reaching selected targets. Has effective weed killing been measured?
Show answer and reasoning
No.
Targeted delivery and a biological control outcome are different endpoints.
The robot operates for every scheduled hour. Does that establish zero crop harm?
Show answer and reasoning
No.
Availability concerns time; crop damage needs its own assessment.

Key ideas

Name the biological and mechanical task endpoints.
Measure desirable actions and unintended crop harm separately.
Include field conditions, downtime, support and safe maintenance in a complete assessment.

References

Sources supporting the explanations and examples in this article.