Three autonomous weeding robots were put through the ringer in Ontario this year – and with good results.
Overall, and despite occasional hiccups, those running the trials said the robots’ performance exceeded expectations.
The tests occurred at a variety of farm sites in Southwestern Ontario, and within a variety of soil types – from marsh muck soil and sand to heavy clay. Haggerty Creek Ltd., an elevator, farm, and ag-service provider initiated the tests.
Autonomy is not a new pursuit for Haggerty Creek and its general manager, Chuck Baresich. Indeed, he and his colleagues have a history of testing new designs – and even marketing some autonomous platforms – including a custom built RoamIO HCT (Haggerty Creek model) from Korechi, OmniPower and OmniDrive from Raven, and others.
The interest prompted the creation of a new company, Haggerty AgRobotics, as well as the establishment of the Autonomous Working Group with equipment manufacturers, the University of Guelph, and Ontario’s provincial agriculture ministry. “We started the working group so we could coordinate our efforts and not have a bunch of different people doing a bunch of the same things,” said Baresich during a presentation given at the 2021 Farms.com Precision Agriculture Conference.
One of the priority projects for 2021 was the testing of several weeding robots. These included two designs from Naïo Technologies, specifically its Oz and Dino models, as well as the GOAT platform from Nexus Robotics.
Smaller than a wheelbarrow, Naïo’s Oz platform operates at slightly under 2 km per hour on an eight-hour operational charge, covering one quarter of an acre each hour. Baresich and his colleagues found its slow speed to be a non issue since it could continue steadily operating – and with “no complaints,” in Baresich’s words.
The Dino is a larger machine capable of covering 10 acres per hour. The GOAT is both larger and slow, since, unlike the Naio products, it employs AI and robotic fingers to pluck weeds both between and within crop rows.
Text continues underneath video
The objective was to demonstrate each machine’s effectiveness within several concepts, in addition to the physical act of weeding, including whether they could operate using GPS alone, the logistical challenges of operating and moving to new locations, effects of heat and dust, and more.
Each machine proved to have advantages and challenges. The GOAT, for example, was capable of removing weeds which could not be hit with the implements attached to Oz and Dino. The small size and tire configuration of Oz, on the other hand, allowed the machine to operate in virtually all field conditions, while the adjustable width of both Niao machines easily adapted to rows of different width.
Some hiccups were experienced, including breakdowns where the source of the breakdown could not be ascertained. However, Baresich says each machine was easy to mend using what are often plug-and-play electric engine parts, plus support from the companies themselves.
Being sent a wide array of surface tillage implements for the Naïo models – many of which were new to the testers – determining the right configuration for the right soil type was also a challenge. Heavily compacted soils and excessive crop residue similarly reduced tillage effectiveness, while tomatoes and other crops with significant canopy growth reduced the time available for weeding. Front-mounted guard plates, though, remedied the latter issue. “Crop growth patterns matter…horizontal crops have huge potential,” says Baresich.
Poor maps equal poor results
Difficulties aside, each robot was able to effectively follow simple A-B lines in the field. “The robot[s] could follow it pretty well. It just needs to know how many reps, and whether to turn right or left,” Baresich says, later reiterating that performance depends on generating good initial maps. The presence of a mistake, that is, cannot be corrected by the machine and will be repeated during each pass. “Poor maps equal poor results.”
The autonomous weeding trials proved about as effective as herbicide-based control plots. In either case, Baresich stresses the importance of controlling weeds when they are small. Repeated passes are also required to achieve clear rows with autonomous units. “Large weeds can plug it up. Uniform soil conditions are ideal, but that’s nothing new,” he says. “RTK is generally good enough to keep the robot away from the crop”
Each of the three robots “exceeded expectations” for those involved. As Baresich and his colleagues move into 2022, they plan to engage in further tests of Nexus Robotics’ GOAT platform. Dino is similarly going to stay, with other yet-to-be-announced technologies slated to arrive later in the year.
Further expansion of the working group is also on Baresich’s agenda, as is the exploration of moving technologies considered field-ready into the market.
In the video below Chuck Baresich talks about Haggerty Creek and why they invest in autonomous technology