A Drone Couldn’t See Soybean Disease Hiding Under the Leaves, So an SIU Team Built a Robot That Rolls Beneath Them

At Southern Illinois University Carbondale, a research team recently learned that the view from above is not always the useful one. The researchers flew a drone over a soybean field infected with a leaf disease and could not collect enough data to detect it. So they built a robot that drives down the rows and looks at the plants from the side and below.

The project is led by Billy Ram, an assistant professor of precision agriculture at SIU, and his doctoral student Samuel Singh. Their long-term goal is AI that spots soybean disease before symptoms show up on leaves and stems, and that farm equipment makers could eventually turn into tractor or sprayer attachments that are efficient, affordable and easy to use. The work, described in an Oct. 13 article by SIU public information associate Brooke Keltner and published by AgriNews, is funded through the Illinois Soybean Center.

The robot is built. The AI that will run it is the next step. That distinction matters, because it would be easy to read a headline about AI disease detection and assume the system already works. According to the article, Ram and Singh are now developing the models the robot needs.

Why early detection matters

Soybeans are a big business in Illinois. The article notes the state is the country’s top soybean producer and says about 43,000 farmers there depend on the crop. It cites the Illinois Soybean Association, which reports that average prices rose from $6.15 a bushel in 2000 to 2004 to $12.71 in 2020 to 2024, while non-land costs climbed as well.

Disease is one of the factors that can wipe out that margin. Many soybean diseases are caused by fungi that live in the soil and spread when it rains, the article explains, and by the time symptoms are visible, the damage is often done. Those fungi can also linger in a field for seasons, which is why the researchers focus on prevention.

“If a disease overpowers your field, late-stage fungicide application is not very efficient,” Ram said. “Once the disease attacks the farmer’s crops, the yield can be reduced to 60% to 70%. It has a real economic impact.”

What the drone missed

The test field at SIU’s University Farms has frogeye leaf spot, a fungal disease of soybeans that causes lesions on the leaves and can spread to stems and pods. Frogeye is caused by the fungus Cercospora sojina and gets its name from the round, gray-centered spots with darker borders that form on infected leaves.

The team started with the tools many precision agriculture projects use first. They flew a drone to make hyperspectral and multispectral maps of the field. Those maps capture wavelengths of light the human eye cannot see, which can reveal stress in plants before a person walking the field would notice it.

It did not work well enough here. Even flying close to the crop, the article says, the drone’s camera had trouble capturing lesions on the undersides of leaves and lower on the plant toward the stem.

“At the end of the day, we’re only working with cameras,” Ram said. “We’re solving an engineering problem more than an agronomic problem. The challenge is how efficiently can you collect a large source of data while the equipment is traveling the field.”

A robot built for the hard angles

The team’s answer was to bring the cameras down to the plants. Their robot is battery-powered and runs on four wheels, with GPS and several cameras mounted on its frame so it can photograph both the tops and undersides of soybean plants. It also has an autonomous mode. A user uploads a map of the field, and the robot follows the rows on its own.

Ram described what it should eventually do. “The robot should be able to drive down the field, keep track of each plant, identify if the plant has a disease and which type and then share what percentage of the crop is diseased,” he said.

For Singh, the end product is a map. “I want to design maps that help farmers detect hot spots, so farmers can see the hot spots and know what site-specific areas to apply fungicides instead of spraying all over the field,” he said.

That goal follows a pattern in much of today’s farm AI. The value is less in a computer naming a disease than in telling a grower where a problem is concentrated, so treatment can be targeted. Spraying only the hot spots can save money on fungicide and reduce how much of it goes onto the field.

Starting from what farmers say they need

Singh, who holds a bachelor’s degree in agricultural engineering and a master’s in remote sensing, has talked with local farmers at SIU agriculture events about the problems they run into. He said that shapes the work.

“As an engineer, we love working on solutions that actually help people, not just something that sounds good on paper,” he said. “The future means humans using robots to make life easier and more efficient.”

The article also notes that SIU recently received the Carnegie Foundation’s Research 1 designation, which recognizes universities with very high research activity. Singh said that status was part of why he chose the program, and Ram said the university is focused on research that will still be useful a decade from now.

What a backyard gardener can take from it

Most readers here will never grow an acre of soybeans, and a desert vegetable bed is a long way from an Illinois field. But the SIU team’s first lesson applies at any scale. The camera that looked straight down missed what was happening underneath.

The same is true of a person standing over a tomato plant or a row of squash. Many leaf problems, and many of the insects that cause them, start on the underside of the leaf or low on the plant, where they are easy to overlook until they spread. Turning over a few leaves and checking near the soil line during a regular walk through the garden is the low-tech version of what this robot was built to do.

There is also a reminder about AI tools in general. Phone apps that identify plant problems are only as good as the photo they receive. A clear picture of the affected spot, including the back of the leaf when that is where the damage is, gives any system, human or machine, a better chance of getting it right.

The SIU robot still has to learn what frogeye leaf spot looks like before it can find it on its own. If the models come together, the payoff the team describes is practical: a field map that shows a farmer exactly where to treat, made by a machine that was built to look where the drone could not.