Engineers are bringing their talents in artificial intelligence and machine learning to the farm.
The Pacific Northwest is home to a number of up-and-coming startups using AI to zap weeds, monitor plant health, and identify rocks in fields.
“The more technology that we can offer farmers to help them with their problems, the better the world is going to be for it,” said Kenny Lee, CEO of Aigen, a Seattle startup that recently unveiled a robot that autonomously thumps out weeds and gathers data for farmers.
Lee met Aigen co-founder Rich Wurden, a former Tesla engineer, in a climate-focused group chat on Slack that helps engineers pivot their careers to address climate issues. They decided to focus on agriculture.
Powered by a low-energy AI model, Aigen’s robot can run on solar power and send real-time crop information to a cloud-based mobile app.
If successful, the startup would be one of the first companies to release a completely autonomous farming-focused machine that does not require charging infrastructure, batteries, or diesel.
Aigen is similar to Carbon Robotics, a Seattle startup that also sells weed-zapping robots. Carbon raised $30 million in April and won the Hardware/Gadget/Robotics of the Year honors at the GeekWire Awards in May. A difference between the two startups is that Aigen is completely renewable powered, Wurden said.
A recent analysis found that there are more than 200 ag-tech AI startups in the U.S.
But ag-tech companies face a number of hurdles in being able to fully take advantage of AI on the farm.
“The biggest challenge in agriculture right now is just getting ground truth data and being able to feed that into a large model,” said Wurden, adding that Aigen sends data to its quantized AI models. “There are no large datasets that are available across the U.S. — or across the world — because that type of data set is incredibly difficult and expensive to collect.”
Aigen’s robots roam close to the surface, collecting data just centimeters from soil and plants, he said. Other companies use drones, IoT devices, and satellite imaging to train their models.
Pollen Systems is a Seattle-area ag-tech startup that uses aerial imagery and individual per-plant data to train its models. It focuses on high-value crops: wine grapes, apples, kiwis, avocados, nuts, citrus fruit, and more.
Pollen Systems uses deep learning combined with visual AI to classify plants — counting them, assessing health, and suggesting actions for various fields through tailored crop profiles for each type.
“This is a work in progress: the more acres of data and imagery that we collect, the better our models get and the smarter our generative AI solutions become,” said Keith McCall, a former Microsoft exec who founded Pollen six years ago.
Another challenge in applying machine learning models to agriculture is increasing a model’s accuracy, said Vivek Nayak, co-founder and vice president of engineering at TerraClear. The Seattle-based startup uses machine learning and hardware to remove rocks from fields.
“We made great progress toward a moderately accurate model,” he said during an ag-tech panel discussion last week at Seattle Tech Week. “But getting from something that’s moderately accurate to highly accurate is extremely challenging.”
He said that companies can try various methods to improve their precision and recall. It’s not just about doubling the amount of inputted data, he said, but also experimenting with different model architectures in computer vision and using strategies such as model sampling and other techniques for higher overall accuracy.
TerraClear also keeps a “human in the loop” for real-time review, Nayak said.
McCall said that AI in agriculture is still in its “early stages.” He envisions personalized AI assistants for farmers, helping them make decisions around water, pesticides, fertilizers, and management techniques based on real-time climate analysis.
Unlike other industries, farmers only have a finite amount of time to perfect methods for growing their crops, McCall said. This highlights the importance of having tools to help them produce crops, he said.
Shifting pressures including rising costs are pushing farmers to be more open to purchasing ag-tech products, according to a McKinsey & Co. report. The study found that 39% of surveyed farmers worldwide intend to adopt at least one ag-tech product within the next two years.
Funding to ag-tech startups fell last year amid the larger tech downturn and have bounced back, but not nearly at the same level as two years ago.
