When John Deere debuted its first-ever autonomous tractor at CES 2022, it signaled a new era of AI & robotic farming would soon be upon us. While other companies have been talking about autonomous tractors for some time, it’s an altogether different matter when the U.S.’s biggest manufacturer of farming equipment signals that this is the future.
Still in the trial phase, early versions of the 8R are now being tested by what the company describes as its “paying test cooperators.” But since it won’t be long before the final production model of the autonomous tractor is rolling off the production line, I thought it would be a good time to sit down with one of the company’s computer vision leads, Chris Padwick, the Director of Computer Vision and Machine Learning at Deere’s Blue River Technology division, to get an idea of the how the company got to this point.
According to Padwick, since its acquisition by John Deere in 2017, Blue River has helped accelerate the farming equipment giant into precision agriculture with its “see and spray” computer vision technology. The technology, which enables a farmer to make highly targeted applications of herbicide to weeds in row crops, first debuted in John Deere’s See and Spray Select in 2021.
The system made it possible to perform green-on-brown application, which is the application of non-residual herbicide to crops in the “pre-emergence” phase. This use of precision application of herbicide allowed farmers to transition away from the blanket application of herbicide to crops using older technology such as cropdusters to more precise application that can reduce the amount of herbicide used by 77% or more.
But it wasn’t until the latest incarnation of the technology, which uses green-on-green technology, that the benefits of Blue River’s investment into deep learning-powered computer vision (which the company began researching in 2016) were fully realized. With the See and Spray Ultimate, farmers can do in-season herbicide spraying for various crops, which is powered by advanced neural network-powered computer vision that can differentiate between similarly colored weeds and crops.
“If plants are touching together, then all of your traditional computer vision techniques for image processing – like morphology and erosion dilation and template matching – kind of break down,” Padwick said. “It’s really not possible to build a system without these that can operate at 95 or above percent accuracy.”
While Blue River helped John Deere accelerate its move into AI-powered farming, Padwick pointed out that the farm equipment company had already invested significantly in the technology even before Blue River’s arrival. By 2019, the company was processing 5 to 15 million measurements per second and had even begun to use computer vision to evaluate grain quality. Much of that work, according to Padwick, was based on work John Deere had performed before Blue River had arrived.
And today, all of the combined competencies and data gathered across John Deere’s various AI efforts are helping the company create its first 8R autonomous tractor.
“In general, in all of our machine learning projects, we tried to embrace the idea that all data is good data,” Padwick said. “We might have sprayer data from See and Spray collected from cameras that can be useful to train the autonomous tractor. The autonomous tractor has different cameras, different geometry, and they can collect data with different kinds of modality and different sensors, but that data that’s collected from other projects can also be useful in that training.”
When asked if he thinks farmers will embrace autonomous tractors, Padwick believes the answer is yes. He points to the rapid acceptance of See and Spray as an illustration of how quickly farmers will adopt new technologies that help them do their job quicker and at a lower cost.
“I remember every time we would do a demo for customers of See and Spray where we’d invite several growers out to a field, and they can watch the machine in action and give us feedback on the results. The overwhelming sentiment was, ‘Wow, I knew this was coming, but I thought it was about ten years away. You guys are showing me the future today.’ So I think the customer sentiment has been fantastic and very hungry for the innovations.”
Padwick says that once early adopters start using innovations like the 8R autonomous tractor, word will spread among farmers, leading to more adoption.
“What I think is going to happen here is you’ll see some people are going to be really excited about the technology and adopting it, and then word of mouth in the coffee shops is going to spread,” Padwick said. “That’s how a lot of these products get sold, not by flashy marketing, presentations, or cool videos on YouTube; it’s coffee shop conversations. And if folks see that other farmers are starting to use the autonomous tractor and getting good value from it, that will naturally drive adoption. Because really, it’s a trust network.”
