When a SprayBox robot identifies a plant, it doesn’t just spray it — it builds a centimeter-by-centimeter map of the entire field, including the geolocation and identification of each plant. (Image: Verdant Robotics)

Unwanted crops and weeds in California’s Central Valley have been on alert — with residents breathing easier — for the last year thanks to Verdant Robotics. That’s because local tractors have been towing the company’s SprayBox technology: AI robots comprising 50 spray nozzles and a sophisticated computer system that aim to empower farmers with high-fidelity information to improve agricultural yields.

Boasting current contracts mainly with carrot farmers but with its eyes set on expansion, SprayBox uses computer vision and machine learning to see objects of interest and act on them.

“You have a bunch of robots behind a tractor, so it’s a very exciting field to work in, no pun intended,” said Georgia Tech Professor Dr. Frank Dellaert, also CTO at Verdant Robotics, located just outside of San Francisco, CA.

Indeed, SprayBox can target individual weeds and crops at a 20-per-second rate and then spray them with de-weeder or fertilizer at a millimeter’s accuracy. In addition, the robots can identify and treat more than 500,000 plants per hour while using 95 percent less chemical weed killer than traditional spraying techniques.

The technology uses AI robots comprising 50 spray nozzles and a sophisticated computer system. (Image: Verdant Robotics)

But that’s not all. When a SprayBox robot identifies a plant, it doesn’t just spray it — it builds a centimeter-by-centimeter map of the entire field, including the geolocation and identification of each plant.

Dellaert noted that SprayBox is not autonomous — there is a person inside the cab and the system is transported behind the tractor. Attached, though, are cameras, which view the crop.

“Then we decide, using machine learning and AI technology, ‘What is a weed?’, ‘What is a carrot?’ And then we have these small sprayers and we spray very precisely on the weeds with an organic chemical,” he said. “And, at the same time, we do research and development that is advanced 3D vision and machine learning. So it’s not rocket science — it’s cooler than rocket science.”

One of the biggest technological hurdles, according to Dellaert, is doing this in the real field. “Mud sprays back to our machines, so we have to be very rugged and figure out how we can get this advanced robotic technology to work for years.”

The other challenge, he said, is applying the 3D technology in the field with live crops and doing the 3D reconstruction.

“There are lots of technical challenges there, such as dealing with the vibration, with the fast movement, and the rig with the 3D nature of these plants,” Dellaert said. “It’s a very cool computer vision and machine learning problem to solve.”

“Since we have all this data about crops and weeds, we can now work with the growers to optimize growing fruit sustainably,” he said. “How can we do this with minimal input of chemicals? How can we figure out maybe from year to year where there is lots of weed pressure. Making this company into an AI company that optimizes agriculture at a much larger scale is our future vision.”

He added that anyone looking to follow suit and implement a new technology needs one thing: a focus on consumers.

“We focus on what our growers needs, which I think is the key,” Dellaert said. “And that applies broadly to anybody who wants to bring technology to a larger market. You need to know who your customers are. And, in our case, our customers are not only growers, but also society. We want to reduce chemical input in the agricultural food-growing process.”

“Currently, we’re still fine-tuning the technology,” said Dellaert. “Our plan is to stay in California for a while to optimize the tech. But we’ve already been talking to growers in other areas, for sure.”

Andrew Corselli is Digital Content Editor at SAE Media Group. For more information, visit here .