The distinction between a weed to be removed and a crop to be preserved in a field has always been a judgment call made by the farmer. This decision is now increasingly being informed by a new artificial intelligence model from Carbon Robotics.

Based in Seattle, Carbon Robotics manufactures the LaserWeeder, a fleet of robots that eliminates weeds with lasers. The company announced its new AI model, called the Large Plant Model (LPM), on Monday. This system can instantly identify plant species, enabling farmers to target new types of weeds without having to retrain the robotic systems.

The LPM has been trained on a dataset of more than 150 million photos and data points. This information was gathered by the company's machines operating on over 100 farms across 15 countries. The model now serves as the core intelligence, named Carbon AI, within the company's autonomous weed-killing robots.

Previously, adapting to a new weed required a lengthy process. According to Mikesell, this took about 24 hours each time. Now, the LPM can learn to identify a new weed instantly, even if it has never encountered that species before.

“The farmer can live in real time and say, ‘Hey, this is a new weed. I want you to kill this,’ and that was something that had never been done before,” Mikesell explained. “There’s no new labeling or retraining because the Large Plant Model understands, at a much deeper level, what it’s looking at and the type of plant.”

Mikesell noted that the company, founded in 2018, began developing this model shortly after shipping its first machines in 2022. He brings years of experience in building neural networks from previous roles at Uber and while working on Meta’s Oculus virtual reality headsets.

This new model will be delivered to the company’s existing robotic systems via a software update. Once installed, farmers can instruct the machine on what to eliminate and what to protect. They do this by selecting from photos the machine has collected, using the robot’s own user interface.

Carbon Robotics has secured more than $185 million in venture capital funding. Its backers include Nvidia NVentures, Bond, and Anthos Capital, among others. Moving forward, the company plans to continue refining the model as its machines in the field supply the LPM with a constant stream of new data.

“We have over 150 million labeled plants now in our training set,” Mikesell said. “We have enough data now that we should be able to look at any picture and decide what kind of plant that is, what species it is, what it’s related to, what its structure is like, without having ever even seen that particular plant before, because we have so much data going into the neural net.”