Carbon AI is trained on 150 million labeled plants, enabling farmers to start laser weeding any field or crop in minutes. | Credit: Carbon Robotics
Carbon Robotics is bringing the era of foundation models to the farm. The company has replaced its collection of crop‑specific AI vision models with a single “large plant model” that understands plant structure across regions and species, allowing farmers to instantly tune Carbon’s laser weeder to their own definition of crops and weeds.
Unlike traditional vision systems that require time‑consuming retraining whenever a new crop, weed, or geography is introduced, Carbon said its large plant model is pre‑trained on millions of images of baby plants collected worldwide and then personalized in the field. Using an iPad app, farmers can simply review thumbnails from their own fields and tag a small number as “crop” or “weed.”
With the tool and the new foundation model, Carbon Robotics said farmers can customize their use of the its autonomous weeder to “zap” the specific plants that they want removed from their field. Even if those plants are desirable plants in another region or part of the farm, the grower can customize the intended target of the laser for each deployment.
Those examples are fed into the foundation model, which immediately adjusts its behavior—without rolling new models or downloading new software. The same global model can selectively zap weeds in a carrot field in Arizona, then move to lettuce or herbs on another farm with only minutes of configuration.
When the LaserWeeder kills weeds, the nutrients from the weed go back into the soil to fertilize the plant. | Source: Carbon Robotics
Carbon Robotics CTO explains crop capabilities
Alex Sergeev, chief technology officer at Carbon Robotics, recently spoke with The Robot Report about the company‘s capabilities.
“What we had to do was to build a process that lets farmers give model examples, and that model doesn’t need time to retrain, just instantaneously work,” he said. “The model can understand differences. So it uses examples, and it’s able to compare those examples to plants that we see in the field at very fast speed. It’s fundamentally different principles.”
Sergeev continued, “The main thing that model needs to output instead of saying, ‘Here’s my confidence that this is a weed’ and ‘Here’s my confidence that this is a crop,’ it gives you a way to do comparison between this plant and all the plants that the farmer told you ‘These are crops’ and ‘These are weeds.’ That’s the difference, and that needs to happen really fast.”
Register now and save on your pass to RoboBusiness 2026iMerit helps to create pretraining data
Labeling data and training the plant foundation model was a huge task. To train its model effectively, Carbon Robotics partnered with iMerit. The data-annotation company delivered the training resources and workflow to ingest the millions of plant images necessary to create a model capable of supporting the new capabilities.
“In the very beginning, we were doing labeling ourselves,” stated Sergeev. “We were very small; we would all go to the farm and then in the evening, we would get together in a hotel room and label images. As you can imagine, it wasn’t scalable. So, in 2020, we went on a quest to find a label partner, and that’s when we found iMerit. We probably did maybe 2,000 images throughout our own kind of labeling life, and they did a million images at this point.”
“We made a tool for our very specific style of labeling, and we just gave that tool to iMerit that was the best because then we could update the tool based on what innovation the learning team made,” he said. “We could also update it based on their feedback or feasibility, so it was like very synergetic that we were able to build this tool quickly with their feedback, and they were able to effectively use it.”
Editor’s note: Sudeep George, chief technology officer at iMerit, recently participated in The Robot Report‘s webinar on “Physical AI and Robotics.”
Carbon ATK is automating the tractor now
Carbon ATK Integrates with smart implements like the LaserWeeder to optimize speed based on weed pressure. | Credit: Carbon Robotics
The next area of innovation for Carbon Robotics is automating the tractor. The company recently released Carbon ATK, an autonomy kit designed to retrofit existing John Deere 6R, 8R, 8RX, 8RT (2019+) series tractors with no permanent modifications.
The Carbon ATK kit includes the cameras and sensors necessary for perception and position tracking. Using the onboard sensors, the unit can perform obstacle avoidance, path planning, tracking, and mission definition.
A tractor outfitted with Carbon ATK and the Carbon Robotics smart implement is fully autonomous and capable of delivering weeding missions without a driver onboard. In addition, the autonomous tractor can be deployed for other farming workflows such as tillage, spraying, and harvesting.
Tell Us What You Think!