Bright
← Living questionsRECORD / Agriculture · Robotics · Energy

Lasers matched the herbicides.

A machine drives a vegetable field, photographs every plant, decides which are weeds, and burns them with short laser pulses. In 2024 field trials in New Jersey and New York, it controlled weeds as well as or better than the standard herbicide programme.

Original sources ↓ · Revision history ↓

Deployed · source published 2025-06-24

What the evidence supports
Verified in the source
A peer-reviewed field trial across peas, spinach and beets in New Jersey and New York found laser weeding left at least 97% less weed biomass, with crop stunting under 1%, matching or beating the standard herbicide programme. An industry-association audit of two Californian farms itemises the cost.
Claimed / not established here
Images processed per hour, laser and camera counts and acreage rates for the machines farmers buy are the manufacturer's figures and have not been independently measured. The trial used a demonstration unit, not a commercial model, and collected no economic data.
Bright editorial interpretation
Acting on one plant at a time turns a field from a single unit to be treated into tens of thousands of separate decisions — a different relationship between farming and chemistry.

Independent support: Sosnoskie, Besançon et al., Pest Management Science, 24 June 2025 (publicly funded; no declared conflicts), with Rutgers Cooperative Extension reporting consistent results and stating that laser weeding is not likely to replace herbicides entirely.

Independent source · Peer-reviewed field trial; at least 97% less weed biomass, crop stunting under 1%, less effective on purslane and annual grasses.

Independent source · Extension guidance: over 97% weed biomass reduction, about 1–3% crop injury, and a caution that it will not replace herbicides entirely.

Independent source · Farm-supplied cost audit: $900.00 to $550.00 an acre combined, and a named crop where the machine cost more than hand weeding.

Primary source · The manufacturer's description of the commercial machine's cameras, lasers and GPU processing.

AI-assisted source review, 12 September 2026. Bright first encountered this work through material curated by NVIDIA, then checked it against the sources above. Other records Bright met the same way ↗

The human problem

Weeding specialty vegetables is done either with chemistry or with people bent double over the rows. Both are running out: 275 weed species have evolved herbicide resistance, and hand weeding in a stooped position has been restricted in California since 2004 — except in organic fields, which have no chemical option at all.

The prior constraint

A sprayer treats the whole bed because it cannot tell a weed from a crop seedling. Telling them apart at speed, plant by plant, is the thing that was missing.

AI’s actual role

Cameras under a light shroud photograph the bed; a deep-learning model classifies each plant as crop or weed, then by type and by growth stage — small, medium or large — and sets the laser's dwell time accordingly. It aims at the meristem, the growing point. The published trial records targeting precision within 1 mm of it. The camera, processor, laser and wavelength specifications are proprietary per the manufacturer, so no independent party can state which chips do the inference.

The documented result

In peer-reviewed trials across peas, spinach and beets at sites in New Jersey and New York, with two to three passes about ten days apart, laser weeding cut weed cover by at least 45% and weed density by at least 66%, leaving at least 97% less weed biomass by season's end. Crop stunting stayed under 1% and crop biomass rose by at least 30%. On two Californian farms audited by Western Growers, hand-weeding cost fell from $900.00 an acre to $282.28, with the machine adding $267.72 — a combined $550.00 and a net saving of about 39%.

Why it may matter

If a machine can act on one plant at a time, a field stops being a single unit to be treated and becomes tens of thousands of individual decisions. That is a different relationship between farming and chemistry, and a different job for the people in the field.

Limitations

The peer-reviewed trial used an autonomous demonstration unit, not a machine a farm can buy, and collected no economic data. Its authors thank Carbon Robotics for donating the unit, though the study was publicly funded and declares no conflicts.

Laser weeding works best on weeds at the two-leaf stage or smaller and loses effectiveness once the growing point is shaded or buried; it was less effective on purslane and annual grasses.

It is slower than a sprayer by roughly an order of magnitude — the economics come from replacing hand labour, not chemistry alone.

On direct-seeded romaine at one audited farm the combined cost exceeded hand weeding, and two competing mechanical cultivators were more cost effective on that crop.

Images processed per hour, laser counts and acreage rates for commercial machines are company figures and have not been independently measured.

Unresolved questions

Does the machine displace hand-weeding work or move it — and what happens to the people who did it?

What is the full footprint once diesel and electronics are counted, against the chemistry it replaces?

Source history & evidence assessment
Maturity
Deployed
Claim confidence
unassessed
Event date
Not recorded
Source published
2025-06-24
Captured
2026-09-09
Last source review
2026-09-12
Editorial method
AI-assisted source review
Place / relevance
New Jersey and New York, United States · study-location

Re-reported from primary sources in September 2026. The headline result is peer-reviewed and publicly funded; the cost figures are farm-supplied and published by an industry association; the specifications for commercial machines are published by the manufacturer and are not independent verification. Bright first encountered this work through material curated by NVIDIA and then researched it independently.

Maturity describes the tested or operational setting. Confidence describes support for the particular claim; one does not determine the other.

Original sources

Deep learning-based laser weed control compared to conventional herbicide application across three vegetable production systems · Pest Management Science · paper

Carbon Robotics LaserWeeder case study · Western Growers Center for Innovation & Technology · report

Laser weeding in vegetable crops · Rutgers Cooperative Extension · institution

Laser Weeding Technology in Cropping Systems: A Comprehensive Review · Agronomy · paper

California Code of Regulations, Title 8, §3456 — hand weeding · government

The International Herbicide-Resistant Weed Database · dataset

Carbon Robotics: LaserWeeder G2 product and mechanism · institution

Institutions: Cornell AgriTech · Rutgers University · Western Growers · Carbon Robotics

Explore the underlying question

Related developments

Editorial connections between distinct settings and results; these links do not imply replication.

More time to decide when to plant.

A field image that asks for less tilling.

A warehouse robot that knows when to stop squeezing.

Revision & correction history

2026-09-09 · Added to Bright as a bounded source claim with visible evidence distinctions.

2026-09-12 · Re-reported from the peer-reviewed trial, the Western Growers audit, Rutgers extension guidance and the California hand-weeding regulation, replacing a record based on a product page.

No corrections recorded.