Carbon Robotics and iMerit enable instant field customization for laser weeders
Carbon Robotics has introduced a large plant foundation model trained on 150 million labeled plants, enabling farmers to customize laser weeding in minutes. The company partnered with iMerit to support this capability. Farmers can tag crops and weeds via an iPad app, and the model adapts instantly without retraining.
The new system replaces a collection of crop-specific vision models with a single foundation model trained on 150 million labeled plant images. Farmers use an iPad app to tag thumbnails from their own fields as crop or weed, and the model instantly adjusts its behavior without new software downloads. CTO Alex Sergeev explained the model performs rapid comparisons between farmer-provided examples and plants observed in the field, rather than outputting static confidence scores.
Carbon Robotics began labeling images internally in 2020, completing roughly 2,000 images before partnering with iMerit. The annotation company has since processed about one million images using a custom labeling tool developed by Carbon Robotics. The collaborative workflow allowed Carbon to refine the tool based on iMerit's feedback, creating a scalable data pipeline for the foundation model's training.
This technology could significantly reduce the expertise and time required to deploy precision weeding, potentially making it accessible to smaller farms that lack specialized AI knowledge. Farmers may gain greater control over their operations, customizing weed removal to their specific conditions. However, the system's reliance on data annotation services and cloud-based models could create dependencies on external providers, and its effectiveness may vary across unusual or highly localized plant species not well represented in the training data.