Robot matches humans in scouting for vineyard diseases

The latest version of an autonomous robot that can scout for grape diseases in vineyards in near-real time, with an accuracy that matches highly trained human scouts, will one day help track crop-killing pathogens with minimal labor.

The robot's capabilities were reported in a paper, "PhytoPatholoBot: Autonomous ground robot for near real‐time disease scouting in the vineyard," published Aug. 25 in the Journal of Field Robotics.

The development of the robot is critical as managing such diseases as powdery and downy mildews in vineyards is the top concern for grape growers and viticulturists.

"It is by far ranked as the highest and most potent threat to the sustainability and viability of viticulture in New York, as well as broadly on the East Coast," said Katie Gold, a grape pathologist and a senior author of the paper, along with colleague Yu Jiang, an applied roboticist, both at Cornell AgriTech in Geneva, New York.

The outlook for grape diseases offers little relief, due to pressures from climate change that favor grape pathogens, regulations that make it harder to get chemicals for treatments, increasing pathogen resistance to fungicides and labor shortages.

While disease is the major concern for viticulturists, declining labor has been the main challenge more broadly within the agricultural food industry for the past two decades, with even worse projections.

"That's been a motivation for us: How we can use robots to do this very skilled job?" said Jiang, assistant professor in the Horticulture Section in the School of Integrative Plant Science (SIPS) in the College of Agriculture and Life Sciences (CALS). "Disease scouting is not something that just anyone can do, [but now our robot] will be able to identify those critical stresses for our food systems."

"In the past, I have regularly hired teams of scouts - four or five people - to sweep the vineyards and do the work of one robot," said Gold, assistant professor and the Susan Ekert Lynch Faculty Fellow in the Plant Pathology and Plant-Microbe Biology Section of the SIPS in CALS. "Now the robot can do it on its own, with just one person babysitting it."

The robot, called a PhytoPatholoBot, can self-navigate between vineyard rows of trellised grape plants.

"While it rolls, it takes side view canopy images and then uses an AI model to infer within the image which pixels belong to the canopy and which pixels indicate disease symptoms," Jiang said.

Information from different image frames is then compared with NASA remote sensing, GPS data and computer modeling that incorporates remote sensing images, to infer disease risk by analyzing the spectral signatures coming off the plants. Plant pathologists and vineyard managers may then receive the robot's calculations in near-real time that reveal the type of disease, locations and infection severity within a vineyard.

The robot not only saves on labor, but also allows growers to target treatments. The information can augment baseline treatment plans by indicating where to target spraying with stronger chemicals, which are in limited supply. "By having an accurate way of knowing where disease is popping up, we could mostly rely on gentler chemicals and only go with the heavy hitters when absolutely necessary," Gold said. Limiting blanket spraying may also help reduce fungicide resistance.

In addition, the ability to gather accurate data on the ground allows scientists to more easily train remote sensing models to passively and more accurately surveil disease.

In the study, the research team deployed and tested the robot across 10 Cornell pathology vineyards, which Gold oversees, as well as commercial vineyards. The PhytoPatholoBot has previously been tested across the U.S. in California, South Dakota, North Dakota, Minnesota, New York and West Virginia. Experimental results have shown the robot's disease detection and severity analyses, while comparable to experienced human scouts and advanced computer vision models, were highly efficient computationally and ran on low power, which is needed for field robots, according to the paper.

A startup in California, co-founded by the paper's first author, Ertai Liu, M.S. '20, Ph.D. '24, a postdoctoral researcher at Cornell Tech, aims to produce the PhytoPatholoBots for commercial use.

The technology is also transferrable to other specialty crops and diseases. "We're really interested in expanding to apples," Gold said.

Coauthors include Lance Cadle-Davidson, adjunct associate professor in SIPS' Plant Pathology and Plant-Microbe Biology Section at Cornell AgriTech; Kathleen Kanaley, a doctoral student in Gold's lab; and David Combs, a research support specialist in Gold's program at Cornell Agritech, who was instrumental in training the robots.

The work is supported by the Cornell AgriTech Venture Fund, the Cornell Institute for Digital Agriculture, the U.S. Department of Agriculture's National Institute of Food and Agriculture, and the NASA Jet Propulsion Laboratory.