Robotic weed management takes to the fields

The potential for cost savings and precise control is enormous

Extraordinary. Revolutionary. Game-changing. Those are just some of the adjectives being used to describe robotic weed management, a technology that is generating considerable anticipation based on the initial research into its benefits in a row-crop environment.

The list of key players and systems includes WEEDit, Xarvio, Bosch, Agrifac and Einböck. And the potential is large enough that major players are entering the sector, like John Deere’s 2017 purchase of Blue River Technology, a company that adapted “machine learning” technology to spray applications.

Related Articles

Last September, Canada’s Outdoor Farm Show hosted a field demonstration of the Dot A-U1 power system. Designed by DOT Technology Corporation, it’s a platform that incorporates seeders, sprayers and spreaders in a completely autonomous, GPS-based operating system. It can house a SeedMaster 30-foot seeder or row-crop planter or a New Leader spreader.

It can also accommodate a Pattison Connect 120-foot sprayer, offering an entry point for robotic weed management. In addition to herbicide applications, the unit can provide sectional control, variable-rate application capabilities and turn compensation.

For Robert Saik, the potential for using robotics in weed management is immense. During the speaking and meeting season, he’s been posing questions to growers and industry stakeholders on what lies ahead for autonomous systems.

One of the questions is about growing crops, and about the struggle to get the right product on the right plant at the right time and at the right rate.

“What would you do if you didn’t have to do it?” asks Saik, chief executive officer of DOT Technology Corp. “When you start thinking about that and start with the whole agronomic process — variable-rate seed, variable-rate fertilizer — the answer there is ‘Yes’ and ‘Yes’. But what about in-season variable-rate fertilizer application as a granular? What about variable-rate herbicide, variable-rate fungicide, variable-rate insecticide, variable-rate desiccation or in-season foliar application or late-season nitrogen on wheat for protein?”

Such questions are getting asked in the followup to the GreenSeeker and WeedSeeker systems of the 2000s. New systems like WEEDit and companies like Xarvio and Bosch are working on optical recognition systems for weeds.

“The technology has advanced far enough where we can identify green-on-brown and spray for it,” says Saik. “That technology exists, but the really exciting stuff is when you start to diagnose and ascertain green-on-green. Could you figure out what sowthistle is inside a soybean field — and kill it? That’s where the technology is evolving.”

At a recent Agrifac meeting in Red Deer, Saik listened to an Australian farmer who’d installed cameras on his sprayer and was experimenting with herbicide applications, reducing them by 90 per cent. Such savings on their own are worth considering, but with Blue River’s latest technology there is a system that eliminates weeds around a lettuce plant, then rogues the weaker lettuce plants. It creates a more uniform, precisely spaced crop, leading to greater consistency of produce.

That’s also where the technology has its easiest, more probable entry point, i.e. into the horticulture sector, where margins are higher than in row crop production and growers can justify the initial outlay and spacing requirements.

Is that spacing issue a problem for robotic weed management in corn, soybeans or wheat? If it’s a robotic tillage implement that runs between plants, then perhaps cereal production would challenge such a system.

Saik foresees another impediment to the adoption of automated weed management and that’s the fact that GPS is not used in the majority of herbicide applications for row crops. Drift in GPS signals can result in applicators driving on the rows instead of between them, crushing crop plants. That, notes Saik, is why driving a sprayer is done manually.

“I sat in a Case sprayer (last fall in Sioux Falls, North Dakota) that was using stereoscopy cameras mounted on the sprayer and it was running between the rows of soybeans, 30-inch rows and 15-inch tires at 15 miles an hour,” he says. “And it was 100 per cent driven by camera-vision guiding the sprayer. The cameras were looking at the rows of the crop and the furrow, and that was guiding the sprayer.”

That technology is currently available and companies are working to bring more of that to the field. It’s one aspect that Saik believes most people don’t realize when considering robotic systems in weed management, yet it’s likely to catch on quicker than any other technology.

Costing the possibilities

The question that enters into most of these discussions is “How much will this cost me?” Last September at the Outdoor Farm Show, the Dot A-U1 platform was listed at US$260,000, and Saik says the pushback on the price has been minimal. In Western Canada, the key determinant is whether a farm can operate more than one Dot and be cost-effective. In Eastern Canada, one Dot platform is expected to fit operations in the 2,000- to 2,500-acre range.

The economic drivers will be numerous, adds Saik, and DOT Technology will be working to compare its systems, including the Pattison Connect sprayer with existing technologies, and the capital expenditure calculations. It’s believed they are $100,000 cheaper than using a high-clearance sprayer.

“The second piece of the puzzle is that operationally, Dot only consumes about 4.7 gallons of diesel fuel per hour,” says Saik. “A third component is that you don’t have a person to drive the sprayer and a fourth is that we believe compaction is less.”

Those factors are to be tested in 2020, particularly the compaction issue since the Dot system weighs 42,000 to 44,000 pounds fully loaded, meaning it’s as much as 12,000 pounds lighter than a high-clearance sprayer.

Yet despite the savings and potential for improving soil health, the biggest challenge with robotic weed management might be convincing a grower that they can trust GPS signalling technology and thus be willing to relinquish some level of control.

“It’ll happen over time, but right now, you can’t blame the marketplace for being skeptical,” says Saik, conceding that progress will be slow, regardless of the positive numbers. “Growers have to see it, they have to know it’s proven so we have a long ways yet. But even by summertime, we anticipate turning heads with Dot doing full-fledged field operations, seeding and spraying. It’s high tech, with high touch, but it’s years in the future.”

More than just driving

Another person who believes in the enormous potential for robotics and advanced weed technology is Mike Cowbrough. He agrees that the capability within these systems and the benefits to growers and the industry are far-reaching. But as with the precision ag sphere of technologies, it often amounts to the same questions: How do I effectively manage my weeds, and what do I use to effectively manage them?

What’s often not being examined in the consideration for these systems and technologies is the physiology of the plants. Herbicides are the chemical means, robotics are the mechanical means, but how will the plant be controlled?

Weed management still requires a human touch, says OMAFRA’s Mike Cowbrough, especially in differentiating between species like redroot pigweed, green pigweed and Palmer amaranth.
photo: Ralph Pearce

“It’s reminiscent of herbicide development and exploration in the 1940s, ’50s and ’60s,” says Cowbrough, field crops weed specialist with the Ontario Ministry of Agriculture, Food and Rural Affairs (OMAFRA). “Ultimately, it’s going to come down to who will be the best at differentiating crop from weed, and once they’re able to do that differentiation, who has the best strategy to kill the weed?”

And that encompasses two components in the robotics sphere: recognizing what is crop and what is weed. The industry is evolving and questioning whether the technology developers have spent too much time worrying about identifying weeds instead of identifying the crop. Cowbrough believes the company that can do that and combine that with an effective strategy to kill the weed will be the most successful.

“My skepticism on the artificial intelligence is that when you’re identifying plant species, you’re not only looking at them, there are lots of senses in the identification phase,” says Cowbrough. He adds there are tactile and sensory aspects in identification. “There’s no question that the technologies are getting cheaper and better and more efficient. But I do think it’s oversold, this idea that a technology can differentiate between common waterhemp and pigweed at the cotyledon stage.”

Yet Cowbrough agrees that perhaps identification of specific weed species is less of a focus than identifying how it grows and reproduces or whether it’s a vining, upright or rosette species. How they’re killed may become the primary concern, in which case the robot still needs the guidance of a person on how to effectively deal with weeds.

“There are a lot of things that conceptually make sense,” says Cowbrough. But, he adds, “There’s this idea that this going to be out in two or three years. Maybe that’s the case, but I think we’re in the early days.”

About the author

CG Production Editor

Ralph Pearce

Comments

explore

Stories from our other publications