Carbon Robotics recently introduced its AI-driven Large Plant Model (LPM), which includes 150 million labeled plants to enable growers to start using the company’s popular LaserWeeder in any field or crop in minutes. 

The LPM, and its AI architecture, is also destined to add precision and convenience to CR’s new Autonomous Tractor Kit (ATK), a retrofit kit designed to enable autonomous operation on existing tractors. Company officials also say the LPM will be used to guide robotic equipment in CR’s forth-coming project introducing AI-trained robotics across a broad spectrum of farm-related operations. 

Paul Mikesell, founder and CEO of Seattle-based CR, says LaserWeeders operating in 15 countries since 2022 have been cataloging various plant/weed species with each field pass, amassing the world’s largest and most diverse agricultural dataset.

“Trained on diverse crops, weeds, soil types, climates, and growth stages worldwide, the LPM creates an unprecedented foundation for plant recognition and decision making,” Mikesell says.

“As the global LaserWeeder fleet operates daily, the system continuously ingests real-world data in a compounding fashion and with our customers’ knowledge uploads that data to strengthen the model exponentially and improve LaserWeeders worldwide,” Mikesell explains. “When our robots can understand any plant in any field immediately and adapt behavior in real-time, farmers immediately get maximum value from the machines.”

Currently, most AI-driven plant identification systems recognize plants/species on which they were trained and throw confidence warnings when certainty of identification is less than 80-90%. Images of those plants are then flagged and inspected by humans who flag the important images, provide ID information and feed the data back into the system to update its future capability.

“That’s a typical AI prediction model pipeline,” Mikesell explains. “Now, we have enough neural net capability to actually be able to decompose the unidentified plants, take thousands of their features and judge them against all the other features of all the other plants the system has experienced.

“That means the LPM can see images of brand new plants it’s never encountered before and not only recognize the plant and what species it is, but be able to see all the variants of it in the future with very little additional data other than a single image.”

He says if a new weed shows up, or if a grower is planting something the system has never experienced, a single picture of the plant is enough to give the LPM sufficient information it needs to do its job without having to communicate back through the cloud to retrain the system.

“That means with a few thumb taps, the farmer can enter the image and the LaserWeeder is equipped to function with the new species information,” Mikesell explains. “We’re able to see new things that our AI has never seen before and re-identify it as long as it’s a plant.

“Say a farmer is working in a field historically not infested with palmer amaranth, but he sees it and consults with our iPad app in the cab, selects a photo of palmer amaranth, taps on it to instruct his LaserWeeder to kill that plant and it will immediately start doing it.”

That capability took cataloging 150 million labeled plants, Mikesell explains.

“We had to get enough data that the neural net was capable of training in a manner different than traditional AI predictive systems. That required a lot of research, but now we’re there and it’s doing a great job.”

Mikesell says the LaserWeeder is in use and has been tested on row crops such as organic corn and soybeans, noting current conventional corn prices do not pencil-out well for adoption of the technology across the Corn Belt yet.

ATK & LPM

Mikesell says CR’s new ATK system, introduced in mid-2025, also benefits from the new LPM because the autonomous system is being used for laser weeding as well as tillage and ground preparation.

“The AI we use for ATK is the same as we use for the LaserWeeder, but we don’t have as complete a catalog of field obstacles for it as we have plant data in the LPM,” he explains. “In the ATK system the AI is running much closer to what a self-driving car would be using. That AI is working well and we’re going to continue to build that library of images to the point ATK can make any tractor completely autonomous for all use cases.”

Mikesell says the ATK world model system sees everything around it and can navigate the field based solely on what it has seen.

“With our system, which can find furrows and follow them — even in fields for which there are no planting or harvesting maps — the ATK can autonomously drive that field without any knowledge of GPS positions.