Weed-plucking robot roams fields

This tiny high-tech robot from Illinois recognizes young weeds and yanks the ravaging intruders hard by the neck 

Herbicide resistance is costing millions. Tillage fosters erosion. So, might you be interested in a 25-pound, 12-inch wide robot that travels between crop rows plucking out weeds?

A new robot with a single-minded focus on yanking young weeds is attracting the attention of Illinois corn growers. TerraSentia is a small, totally independent robot that works for 10 consecutive hours on a battery charge. It can’t work at night because it requires daylight for the sensors to accurately discern the difference between crop and weed.

The weed machine started life as a plant-breeding tool to help researchers in their high throughput phenotyping projects in corn and soybeans. The robot was initially designed to prowl up and down the rows in trial plots of experimental corn and soybean varieties, said University of Illinois researcher Girish Chowdhary, project head on the multi-disciplinary TerraSentia program.

The weed yanker can be programmed to identify and kill herbicide-resistant weeds in-crop for a projected cost of less than $100 per acre when it gets to market. The cost of $100 per acre may not be so outrageous if you consider that your three options are intensive tillage, simply living with herbicide-resistant weeds or hand roguing.

In many parts of the world including North America, producers with resistant weeds have resorted to hoeing and hand roguing in small fields with high value crops like potatoes or sugar beets. But that’s not practical with broad-acre crops. Besides, how many people will you find today to hand rogue or hoe a field — and at what cost?

On a personal note, this writer hoed sugar beets for three summers while in junior high. One summer a grower waited too long to call us “hoe kids” and the weeds took over. For the next week we were bending over hand-yanking weeds by the roots — for quadruple the normal pay. I’m sure the grower would have been happy to have spent only $100 per acre. No doubt that whole experience convinced me to shape up in school and get into university.

In a phone interview, Chowdhary explained that for plant-breeding purposes, they designed sensors to collect data on numerous factors, including plant height, biomass, corn ear height, leaf area index, disease resistance and early plant vigour. Armed with this data, plant breeders compare each genetic line to other lines in the trials. Before automated, high volume phenotyping, this data was gathered manually with human observation. Chowdhary said TerraSentia makes far better observations than the human eye.

If TerraSentia can autonomously roam a corn field picking out the difference between crop and weed, the next step is to weaponize it to destroy the weeds one by one.

Once the team saw how quickly and precisely their prototype machines identified weeds, it was clear they had a device with great potential. It can also be an important tool for organic farmers who don’t want to cultivate because they don’t want to create soil erosion.

Chowdhary said the phenotyping work continues and his team is now concentrating on developing a mechanical de-weeding robot.

“The prospects are immense for using this device in fighting herbicide-resistant weeds. More herbicides won’t solve our problem. Tractor-pulled equipment only works in some crops and some parts of the season,” he said.

“We think we can succeed in breaking the herbicide-resistance problem by using a large number of small robots that can travel between the rows killing all young weeds and keeping those rows clean until we have shade from the canopy.

“We’re in Illinois, so of course we’re working first on corn and soybeans. So far, it works better in corn than in soybeans.

He said he thinks the system can eventually work on cereals, canola and other broad-acre crops, but their effectiveness depends on row spacing.

“What I’ve seen with your tight row spacing in some Saskatchewan canola fields is the tractor tires or the sprayer tires drive over a lot of the vegetation and that hurts your yield. Currently, the way you grow canola in Saskatchewan, I don’t think a robot can work. But I’m sure we can build a smaller unit that will work in 15-inch row spacing as some of your farmers now use.”

Now in its 18th prototype version, Chowdhary said his team had 30 units working in corn and soybean fields this past summer.

The little robots are built with 3D printers. He said they work well in row spacings of 25 to 30 inches. As each robot travels between plants, it also moves laterally left to right, creating a clean sweep of the surface and even peering into the gaps between the plants. Once the robot is in production, it will also be able to extricate weeds growing in the seed row.

In early season when the crop is still small, TerraSentia recognizes the difference between crop and weed with an error of less than four percent. Later in the season, when the crop is taller, the error is closer to nine percent.

Chowdhary said designing a smaller TerraSentia is not a major obstacle. A bigger challenge for his engineers is how to improve the optics so it can keep working longer into the growing season when the crop shades out necessary daylight. Once they’ve conquered that issue, robotic weed control might extend far enough into the growing season that herbicide use can be greatly reduced or even eliminated.

Driving any kind of four-wheel vehicle on a soil surface eats up energy. He said the robot may only weigh 20 or 25 pounds, but it was still a challenge to find the best power-to-weight ratio for the lowest draw on the battery. The controller continually tunes the drive system to keep the robot moving while using the least possible amount of power.

The team is evaluating a number of technologies for weed extermination. The obvious one is a simple pair of fingers that pull a weed out by the roots. They’re also considering brushes and knives that undercut the weed.

In parallel research, a number of departments at the university are working on sophisticated robots that can identify an apple on a tree and pick it, or even pick an individual berry from a bush. Chowdhary said the dexterity of these new hands and fingers can easily be applied to the TerraSentia weed-kill project.

“We hope to bring this to market for a cost less than $100 per acre per season. A farmer has to calculate what is his loss per year to herbicide resistance. A different way to calculate is to ask how much you would pay per acre to get herbicide resistant weeds out of your fields.”

Manufacturing and marketing of TerraSentia will be handled by a new company called EarthSense.

About the author


Stories from our other publications