The GPS that directs hulking pieces of machinery along precise and invisible rows can’t be relied on at the smallest scale — where compact robots roam under satellite-obscuring stalks and leaves.

Despite these limitations, a new wave of autonomous robot workers for the agriculture industry is on the horizon, and engineers at the University of Illinois are leading the way.

A lab at the Urbana campus of the University of Illinois, led by assistant professor Girish Chowdhary, is building 3-D-printed robots to collect data on individual crops. Over years of prototyping, these robots have gotten skilled at using code and sensors to differentiate individual crops from background noise and weeds.

Additionally, the latest robot isn’t stuck to moving in a straight line — it effectively draws its own line by looking at its surroundings.

“This robot is running around; it’s detecting the corn and it’s tracking it,” says Chowdhary of the 18th prototype. “Early emergence, we get pretty good accuracy, with a 3.8% error. So early season is good. But late season has more challenges. Our accuracy is about 91%, and improving.”

Chowdhary and his lab are in the final year of an Advanced Research Projects Agency-Energy-funded project under the TERRA program that helped them develop this robot. TERRA stands for Transportation Energy Resources From Renewable Agriculture.

Now, Champaign, Ill.,-based startup EarthSense is taking the robot named TerraSentia and selling it to early adopters interested in gathering data about early plant vigor, plant height, corn ear height, leaf area index, biomass and disease pressure.

“The TERRA project was all about measuring plant traits in a quantifiable, repeatable way and autonomously,” Chowdhary says. “With that, we matured the robots — we matured the machine vision algorithms and navigation algorithms.”

Now that the robot can accurately differentiate between crops and weeds as it moves autonomously through fields, Chowdhary and his lab are building out physical weed-fighting capabilities for the robot under new National Science Foundation and National Institute of Food and Agriculture grants.

“We’re looking into how to make it practical and feasible to mechanically de-weed with these robots,” Chowdhary says, adding that his work is motivated by a growing herbicide-resistant weed problem in agriculture that is costing billions of dollars each year in lost yields.

While hard, mechanical arms can do the trick, Chowdhary says his lab and other experts at the university are working on developing octopus arm- and elephant trunk-inspired “soft arms” to tackle engineering challenges such as reaching behind a leaf to grab an apple.

“The next generation of problems is when you have to grab a berry. Or look behind a leaf. How do we do those dexterous things with an arm that was originally designed to drill a hole or hold screws?” he asks.

Chowdhary has worked with airborne drones since 2003. He says he came to ground-level robots in part because if they’re small enough, they can noninvasively work under the canopy. He notes that the robot currently works better in cornfields than in soybean fields.

“In the Midwest, we’re lucky in the sense that if you’re conventional farming, you can use large equipment and make a decent amount of money. But if you’re an organic farmer, or if you’re trying to do a labor-intensive crop, you’re kind of out of luck. We want to help change that,” Chowdhary concludes.