WEST FARGO, N.D. — For now, they informally call it a “weedbot.”

It’s a remote-controlled robot on wheels that scientists hope will eventually will use artificial intelligence to spot-spray individual weeds in crop fields — remotely.

Xin (pronounced SHIN) “Rex” Sun (SOON), a North Dakota State University Agricultural and Biosystems Engineering assistant professor, is co-principal investigator on the federally-financed project at the university.

The project for now is for research and education but ultimately, principles learned today could help tomorrow’s farmers use robots to control weeds in the field.

Farmers got a look at the futuristic project on Sept. 13, 2022, at the Big Iron farm show in West Fargo. The machine is there through Sept. 15, 2022, from 1 p.m. to 3 p.m., in front of the Hartl Building.

Sun and colleagues are working on the project in steps.

The first step is weed identification data in greenhouses. Using track on top of the greenhouse bench, they’ve made more than 5 million photographic images that are used to “recognize” important weed species. That goal is to identify weeds when they are immature — when they’re exceedingly difficult to recognize visually, but most vulnerable to spray control..

“By the time they’re 4 to 6 inches tall, you don’t need any artificial technology” to identify weeds, Sun said. “You want to recognize them when they’re small, like little tiny dots outside of the soil. That’s the challenging part. But I’m working on that.”

The idea is to spot spray weeds, economically, before they’re big enough to hurt the crop. As the technology progresses it will also be easier on the environment.

Ultimately, other equipment on the same machine will use artificial intelligence and its own spraying system to individually, remotely take out the particular weeds.

The first research step in a four-year project was announced in 2018. Sun and colleagues worked in a greenhouse, running camera systems across populations of young weeds and crops.

They use four cameras — multi-spectral (four bands of light), hyper-sectral (hundreds of waves), thermal sensors and RGB (red, green, blue) — and from various angles. The scientists expect to continue to collect those for about another year.

Within the five million photos are the crop species — flax, edible beans, lentils, sugarbeets, soybeans, wheat and corn. They also photograph several primary species of weeds.

The goal is to try to define which camera combination works for which crop or weed species.

Then they installed a similar system on their field version of the contraption. Some differences between weeds and crops cannot be seen with the human eye.

“Version 1” had wheel, aluminum frame, measuring about six feet tall, and six feet wide. Four people can lift it.

Sun and colleagues started testing that machine at NDSU Agronomy Seed Farm at Casselton and then demonstrated it at Grand Farm, south of Fargo. Grand Farm is a non-profit enterprise that allows researchers to collaborate. Sun and colleagues also used their robot at NDSU's Carrington Research Extension Center. They created Version 2 and Version 3, and now Version 4.

It is the “Version 2” that was shown at Big iron. This is the same remote-controlled frame as Version 1, with a customized spray tank, camera system and a computer vision system. It is remote-controlled and is not run with GPS (global positioning system satellites) as of yet.

The two-wheel-drive system has a set of batteries and an electric motor.

A boom in front is equipped with a high-resolution camera that takes pictures that “detect” plants (crops and weeds), creating a high-resolution data base.

The machine is equipped in the middle with a “low-resolution but fast-processing” camera that signals a control box that has an artificial intelligence processor. This system gives the signal to spray, from a 50-wide boom, which currently has three nozzles for demonstration purposes.

“It is still not ‘autonomous,’ but remote-controlled,” Sun said.

So far, Version 2 is just shooting water from a 25-gallon spray tank. Ultimately, it will use herbicides.

“When you involve the chemical (herbicide) there are regulations,” Sun said. “We’re spraying with the water just to prove the concept.”

Sun knows of some of the other research institutions and companies working to solve various aspects of the same robotic problems. John Deere has a “See & Spray” technology. The technology is “already here” but is being customized.

He acknowledged he’s primarily is focused on his own system, “working on North Dakota soil, dedicated to North Dakota agriculture,” as he’s fond of saying.

Sun is coordinating with weed scientist Kirk Howatt, an NDSU associate professor of plant sciences, and Tom Peters, an NDSU and University of Minnesota Extension weed specialist.

Howatt helped identify which weeds should be studied in the project — kochia, ragweed, horseweed and redroot pigweed. They also study the dreaded Palmer amaranth weed, but cannot transplant it into the outdoor field tests, because of worry it could unintentionally spread. “That’s too dangerous,” Sun said, gravely.

Howatt helps Sun grow weed samples in the greenhouse. Project technicians successfully, safely transplanted for study, so they can be “detected” by the weedbot.

As the equipment is trained to improve weed control, the goal is future versions.

(For starters, they’ll probably formally call it something other than “weedbot,” a name used by a a for-profit company, WeedBot , founded in March 2020, based in Riga, Riga, Latvia.)

They’ll probably task future versions to engage with diseases and other pests.

“You can use the same camera and change the direction the camera is pointing to,” he said. “Instead of pointing down, you can change it to point to the side of canopies. The same robot, vehicle, equipment, can detect early signs of disease for corn, wheat, or different crops.”

Version 4 was built by 20 senior design students in NDSU’s Agricultural and Biosciences Engineering department, led by Tom Bon and Brian Gregor. It's called a “multi-functional robotic platform.” It is eight feet wide, 12 feet long and driven by 67-horsepower engine. And this big fellow potentially could include automatic collection of soil samples or even smart weather station. That one is not ready for a prime time viewing, Sun said.

Maybe they’ll show that big guy next year at Big Iron.

Sun and his colleagues must keep looking ahead.