The determination of what qualifies as a weed in the field and needs to be eradicated is traditionally left to the discernment of farmers. However, Carbon Robotics is ushering in a new era with the introduction of an AI model that is set to transform this process.

Based in Seattle, Carbon Robotics is the creator of the innovative LaserWeeder, a fleet of robots equipped with lasers designed to effectively eliminate weeds. Recently, the company unveiled the Large Plant Model (LPM), a groundbreaking AI model that can instantly identify plant species, empowering farmers to target new weeds without the need for extensive retraining of the robots.

The LPM has been meticulously trained on a vast dataset comprising over 150 million photos and data points collected by Carbon Robotics’ machines from more than 100 farms across 15 countries where the robots are currently operational. This model now serves as the core component of Carbon AI, the intelligence system that drives the company’s autonomous weed-killing robots.

Paul Mikesell, the visionary founder and CEO of Carbon Robotics, shared with TechCrunch the significant impact of the LPM. Previously, whenever a new type of weed appeared on a farm or variations of existing weeds emerged, the company had to devote considerable time and resources to create new data labels for retraining the machines to recognize the plants. This process typically took around 24 hours for each new weed. However, with the advent of LPM, the identification of a new weed is instantaneous, even if it has never been encountered before.

Mikesell emphasized the real-time capabilities of LPM, highlighting how farmers can now promptly alert the robots to a new weed and request its elimination without the need for time-consuming retraining. This seamless integration of the Large Plant Model represents a monumental leap in efficiency and accuracy for Carbon Robotics.

The journey towards developing the LPM began shortly after Carbon Robotics initiated the deployment of its first machines in 2022. Drawing on his extensive experience in neural network development from previous roles at Uber and Meta’s Oculus, Mikesell spearheaded the creation of this cutting-edge model.

The seamless integration of the LPM into Carbon Robotics’ existing systems will be facilitated through a software update. Farmers will have the ability to instruct the machines on which plants to eliminate and which to protect by selecting relevant photos within the robot’s user interface.

Having secured over $185 million in venture capital funding from prominent investors such as Nvidia NVentures, Bond, and Anthos Capital, Carbon Robotics is well-positioned to further enhance the LPM. Continuous refinement of the model will be driven by the influx of new data from the robots as they operate in diverse agricultural settings.

Mikesell expressed confidence in the capabilities of the LPM, citing the extensive dataset of over 150 million labeled plants that underpin its functionality. The neural network’s robust training enables it to accurately identify plant species, analyze plant structures, and make informed decisions without prior exposure to specific plants.