Weeding robots from plant location to verified weed control

Two documented systems use different information

Carbon Robotics' original LaserWeeder is a tractor-mounted implement that uses cameras and lasers. The manufacturer distinguishes it from its earlier self-driving prototype. Automatic weed treatment should therefore not be read as a claim that this implement drives itself between fields. [1]

FarmDroid's mechanical system instead uses recorded seed positions from its own seeding operation. It describes weeding between rows and within rows using that position information, without relying on plant-recognition cameras for this process. Compatibility with the planting workflow is part of the system. [2]

Scroll sideways for all columns.

Tool position must match the plant when treatment occurs

Consider an illustrative machine moving at 0.5 m/s. A 100 ms delay between the relevant observation and treatment corresponds to 5 cm of travel. This calculation shows why timestamps and motion compensation matter. It is not a measured delay or accuracy figure for either product.

Use the actual treatment point in the evaluation. Correct classification on a stored image does not establish that a moving tool contacted the intended location later. Likewise, a good planting map needs to remain aligned with the implement's physical coordinates.

Have the supplier document allowed operating conditions and the approved procedure for calibration. Do not improvise tests around exposed laser radiation or moving agricultural tools.

Count living weeds after treatment

An illustrative plot starts with 200 target weeds. The system treats 180 and a follow-up inspection finds that 150 of the original weeds have been controlled according to the agreed rule. The treatment coverage is 90%. Observed weed control over the original population is 75%.

Define the follow-up interval with an agronomist and state it in the report. An immediate image of a treated point cannot replace a later assessment when the chosen method requires one. The example's numbers are not a product benchmark.

Keep crop injury in a separate column

Suppose the same hypothetical plot contains 1,000 crop plants and eight show treatment-related injury under the trial's assessment rule. Report the 0.8% figure separately from weed control. A high weed-control percentage should not hide the effect on the crop.

Also record untreated crop-adjacent weeds, weeds in headlands and areas the robot skipped. Excluding those regions after the trial begins can make output look better than whole-field performance.

Compare methods on the crop species, growth stage and field arrangement where they will be used. The relevant tradeoff can include plant damage, repeat passes and operator work as well as time per hectare.

Plan for the tractor, tools and remaining work

For an implement, include the towing or carrying vehicle, its driver or control system and transport between fields. For a self-propelled system, establish the boundary of autonomous operation and how people handle exceptions.

The field log should include productive treatment, turns, cleaning, adjustments and stoppages. A useful contract identifies the area and quality outcome to be delivered. It does not rely only on a count of laser shots or mechanical passes.