I, Growbot: How Robot Weeders Are Transforming Agriculture

Now there is innovative AI and machine learning technology to assist farmers with weeding their crops, bringing a whole new meaning to seed funding.

By Danny Bradbury

Robots have affected manufacturing and medicine, and have even explored the surface of Mars. Now companies are hoping that they can take on something far more mundane: the common weed. Robo-weeders are tidying up everything from domestic gardens to commercial vegetable fields, and the results could have a serious impact.

When the co-designer of the Roomba robo-vacuum becomes interested in robotic weeders, the rest of us should take notice. After steering that product to success as chairman and  president of iRobot Corp, Helen Greiner is now CEO of Boston, MA-based Tertill Corporation , a startup that began as a Kickstarter project. Its sole product, the Tertill, is a small garden robot that takes a simple approach to weeding.

“Over my career in robotics, it’s the companies that think of a different way to do things with robots that have the big successes,” says Greiner, explaining that she drew on her experience at iRobot to create a device that wasn’t driven by machine learning or computer vision, but instead used simple sensors to spot garden weeds. “The Tertill was in the top 1 percent of technology Kickstarters.”

“Over my career in robotics, it’s the companies that think of a different way to do things with robots that have the big successes.”

–Helen Greiner, CEO, Tertill

Tertill assumes that large plants are wanted, while shoots shorter than an inch are emerging weeds. It scours garden plots, using a dozen sensors to discern between the two. It uses a nylon strimmer to cut down small weeds without removing them, sapping their energy while also returning nutrients to the soil. Gardeners can also insert simple metal hooks into the ground to protect single small plants and rows of seedlings.

Greiner targets vegetable gardeners with the Tertill, which is already shipping to consumers. She says it appeals to customers who don’t want the hassle of weeding, but who shy away from herbicides. Some small organic farms have also adopted the unit.

“Those folks are going to have a much easier time trying it out at a consumer price point,” Greiner explains. “They can just buy one for $349 and try it out on a section of their small farm to see how it looks rather than investing $20,000 in a piece of capital equipment that they don’t know is going to work for them.”

From Gardens to Farms

Robotic weeders could be more than just labor-saving gadgets for consumers and small businesses. Salinas, California-based FarmWise sells an industrial-scale robotic weeder that uses artificial intelligence (AI)-powered computer vision to help vegetable farms increase their crop yield while coping with labor shortages and dwindling herbicide options. It plans to be a lifesaver for commercial farming operations that face a shortage of people to handle traditional manual vegetable weeding tasks. “Labor isn’t cheap, and it’s not easy to find, and everybody fights for the same kind of people,” says the company’s business strategist Pauline Canteneur.

While the U.S. government subsidizes some commodity crops such as corn, wheat, and soybeans, it doesn’t subsidize specialty crops, which include vegetables, according to the National Agricultural Law Center. Yet growing costs for vegetables are inordinately high. The total operating costs for wheat stand at $534 per acre, according to research at the University of California. That jumps to $11,500 for lettuce thanks to more intensive costs at both the growing and harvesting stage. These include insect and disease control, hand weeding, and field packing. Unlike commodity crops that can be fertilized and harvested in bulk using mechanized equipment, a lot of vegetable farming is still manually intensive. Robotic weeders won’t diminish all those costs, but FarmWise aims to help.

FarmWise’s AI-powered equipment recognizes the 3D geometry of each plant to understand where its stem and leaves are located, explains Canteneur. After recognizing the crops, it uses blades to uncover the roots of everything else around it in the vegetable row, exposing them to sunlight so that they die out. This is a more sophisticated version of Tertill’s leave-it-in-place approach, returning dead weeds as nutrients to the soil.

FarmWise’s AI-powered equipment recognizes the 3D geometry of each plant to understand where its stem and leaves are located.

The company began by using its weeders to take pictures of crops which technicians then annotate and use to train a sophisticated AI model. As it deploys its robots in the field, they continue to image crops as they weed them, allowing FarmWise to refine its model still further. It can take up to two months to train the software sufficiently for a crop, and currently covers row-based crops such as lettuce, brussels sprouts, artichokes, and cabbage. FarmWise already has over 10 machines working on farms in Salinas and is on its third generation of machines, which uses an autonomous, diesel-powered tractor that pulls the weeder behind.

Weeding as a Service

One big problem today is Californian regulations that insist on operators being no more than 10 feet from an agricultural machine. That’s a problem when trying to scale an autonomous weeder business, and means that FarmWise must always have one of its own operators on-site. “The law that asks us to have an operator is 6 years old now. It needs to be changed, because technology changes at a faster pace,” Canteneur says.

Regulations like these are less significant north of the border in Quebec, where Nexus Robotics has its own unique approach to robo-weeding. Like FarmWise, its software uses supervised machine learning to discern crops from weeds, but instead of using blades to expose weed roots, its AI-powered robot arms pluck the weeds from the ground.

“Pulling weeds means that we can go extremely close to the plant,” says Nexus CEO Luc Labbe. “We can pick any and all weeds.” He claims a 95 percent weed-picking rate with the second-generation unit, which can pick a weed every two seconds on average, and plans to deploy a third-generation unit this year that will be faster still.

In addition to three large farms in Quebec, the company hopes to use the autonomous third-generation weeder in Nova Scotia and Florida, where regulations will permit it to operate without any staff on-site.

These machines are expensive enough that both FarmWise and Nexus Robotics had to rethink traditional agricultural business models. They both sell “weeding-as-a-service” contracts, in which they deploy and look after the machines, charging by the acre.

“We see it as a way to bring a sophisticated new technology to the field, right now, instead of waiting five years until the product gets really mature,” Canteneur says. That way, the farmer gets a quality service while the AI company gets to train and refine its product, and gets paid. In a nascent field still weeding out early technical challenges, it gives a whole new meaning to the term ‘seed funding.'”