Carbon Robotics LaserWeeder G2 Explained: How It Tells Lettuce From Weeds

How LaserWeeder G2 sees, thinks, and fires

Step one: make the plants visible

Computer vision begins with light. A camera cannot classify a leaf edge or growth point that it cannot see clearly. Carbon lists 20 LED lights inside each G2 weeding module and describes operation during the day or night. The lights create a more controlled view under the implement while the surrounding field changes from bright sun to shadow.

Each module is listed with three high-resolution cameras. Their job is to turn a moving crop row into a stream of images. A useful image must preserve enough detail to separate leaves, stems, soil, residue, overlapping plants, and the target area while the tractor is moving.

This is why the machine’s large frame is only part of the story. The visually interesting work happens close to the soil, where lighting, optics, vibration, dust, mud, plant motion, and timing all affect what reaches the model.

Step two: decide which plant belongs

Carbon says the camera feed goes to onboard computer-vision models that identify crops and weeds. The current G2 page lists two NVIDIA GPUs per module and more than 100 crop models. The GPU does not know what lettuce is in the human sense. It performs many calculations on the image and produces a classification and target location based on learned patterns.

Carbon’s current AI page describes a Large Plant Model trained on 150 million labeled plants collected across 15 countries, with support language covering more than 100 crops. Those are company-stated dataset and coverage figures, not a BTI audit of the labels or a guarantee for every field.

The page also describes Plant Profiles. An operator can select two or three photos to define local crop and weed conditions, after which the system adjusts model behavior for that field. That detail matters because the same species can look different by growth stage, variety, soil, weather, camera angle, or damage. A broad model supplies a starting point; the field profile tells the system what the operator wants it to protect and target now.

Step three: aim at the growth point, not the whole plant

After classification, the system needs a coordinate. Carbon describes its laser as applying thermal energy to the weed’s meristem. The meristem is growth tissue where new plant cells form. Damaging that small area can stop regrowth without pulling the plant or disturbing the soil around it.

That is the beginner-friendly reason precision matters. The machine is not trying to burn an entire patch of green. It is trying to place energy on a selected point inside a plant that has already been classified as a weed. If the classification or target location is wrong, precision alone does not fix the decision. Classification and targeting must both be right.

The G2 page uses sub-millimeter detection language. Treat that as a Carbon specification. Ask for the measurement method, crop conditions, confidence thresholds, miss rate, crop-injury rate, and what the operator sees when confidence is low.

Step four: apply heat with a sealed laser module

Each G2 module is listed with two 240W lasers. Carbon describes the module as self-contained and fully sealed, with liquid cooling and a 21-inch shoot width. The energy is applied to the selected weed target while the implement travels through the field.

The published safety label is important: LaserWeeder G2 is a Class 4 laser product with invisible laser radiation. A Class 4 label is not a decorative technical detail. Direct or scattered exposure can be hazardous to eyes or skin. The equipment’s physical enclosure, interlocks, operating procedures, maintenance controls, exclusion zones, training, and manufacturer instructions are therefore part of the product, not optional extras.

BTI is not providing laser-safety certification or operating instructions. A prospective operator should obtain the current manual, site requirements, training plan, service procedures, and applicable workplace rules directly from Carbon Robotics and qualified safety personnel before use.

What one complete G2 400 configuration contains

Per-module numbers can feel abstract, so Carbon’s G2 400 page provides a useful named example. It lists eight modules, 16 x 240W diode lasers, 24 high-resolution cameras, and 160 LED lights. It also lists more than 100 AI crop models and a maximum of 6,667 weeds shot per minute.

Those figures describe the G2 400 configuration on the manufacturer page. They are not independent BTI measurements, and the maximum weeds-per-minute figure is not the same as a promised acres-per-hour result in every field. Carbon lists a coverage range of 0.80 to 1.60 acres per hour for that model, which will depend on the operating setup and conditions.

The broader G2 family runs from 8-foot to 40-foot models. Carbon labels them G2 200, 300, 400, 600, and 1200. More width and modules can cover more rows, but they also change tractor, lift, power, transport, field-layout, service, and capital requirements. This is industrial equipment sold through contact sales, not a consumer product with a simple add-to-cart comparison.

A practical evaluation checklist

Start with the exact crop, variety, row spacing, soil type, weed mix, growth stages, weather, and tractor that will be used. Request an uncut field pass rather than a short highlight clip. Mark sample areas before the pass and count crop plants, weeds treated, weeds missed, and any visible crop injury afterward.

Review the operating workflow: who selects the crop profile, who can change targeting parameters, how software updates are controlled, what data leaves the farm, how a machine is stopped, and what happens after a sensor, network, cooling, or laser fault. Ask how service access and remote support are logged.

Finally, build the economics from local evidence. Use the quoted machine and service terms, expected annual acres, tractor and operator time, maintenance, transport, energy, current weed-control costs, crop value, and observed field results. Carbon publishes commercial benefit claims on its site, but BTI has not verified a payback period for a particular farm.