How John Deere Grows Dataseed into an AI Powerhouse

During CES2022 in January, John Deere debuted a fully autonomous tractor with artificial intelligence. It is ready for large-scale production.

According to the press release, the tractor has six sets of stereo cameras that capture images and pass them through deep neural networks. This classifies each pixel in about 100 milliseconds and determines whether the machine will continue to move or stop. An obstacle has been detected.

And in March, the Iowa-based company launched See & Spray Ultimate, a precision-targeted herbicide spray technology designed by John Deere’s wholly owned subsidiary, Blue River Technology. Cameras and processors use computer vision and machine learning to detect weeds from crops. One camera is mounted every meter across the width of the 120-foot carbon fiber truss-style boom, or 36 cameras scan more than 2,100 square feet at a time.

But John Deere’s position as a leader in AI innovation hasn’t emerged out of nowhere. In fact, agricultural machinery companies have been planting and growing data seeds for over 20 years. Julian Sanchez, new technology director at John Deere, has invested heavily in the development of data platforms and machine connectivity, as well as GPS-based guidance, over the past 10-15 years.

“These three elements are important for AI conversations, as the actual implementation of AI solutions is mostly data games,” he said. “How do you collect the data? How do you transfer the data? How do you train the data? How do you deploy the data?”

Recently, the company has enjoyed the fruits of AI labor, and the yield will increase further in the future.

John Deere’s long journey to AI

John Deere’s efforts in developing artificial intelligence solutions are part of a larger trend across agricultural landscapes. According to Markets & Markets, spending on agricultural AI technologies and solutions is projected to increase from $ 1 billion in 2020 to $ 4 billion in 2026.

The company’s path to AI began in the mid-’90s when a small group of innovative engineers were told to move away from John Deere’s product lines, such as the Harvest Combine Group and Tractor Group, to Des Moines, Iowa. I did. We are working on a wave of new technologies centered on GPS.

According to Sanchez, the GPS-based steering system released in 1999 was a turning point in tractor accuracy at John Deere. “With less overlap, the economics of accuracy are easy to understand,” he said. “But what pitched the farmers was not whether they stayed in a straight line, but the ability to monitor other parts of the work. It was a big unlock. Since then, we’ve been building on it. I did. ”

For AI opportunities

Sanchez explained that the next “Aha” moment was when John Deere tagged all sensors in the vehicle with geospatial positions. “Sensors are associated with all kinds of agricultural work, such as sowing seeds on the ground, harvesting plants, and spraying herbicides, so some are working well in the field and some are not. I see, “he explained.

This opened up the whole idea of ​​geospatial maps, and John Deere began development immediately in the early 2000s. However, data transfer was awkward. “They were recorded on the machine, so I had to use a USB drive to collect them all and put them back in the farm and upload them to my PC.”

As a result, in 2010, John Deere realized that all large agricultural vehicles shipped from the factory would need to come with a cellular-enabled telematics box. “We have begun to remove the friction of having to move data from the vehicle to another location in order to move the data continuously from the vehicle,” he said.

The 2010s revolutionized mobile and cloud, accelerating the innovation capabilities of digital tools. By 2016, Moore’s Law (the principle that computers can be expected to double in speed and functionality every two years) has revived the opportunities for AI to do. At that time, John Deere had several small teams working on the concept of robotics for at least 10 years. “We have worked with some of the top robotic universities in the country,” Sanchez said. “In other words, we can essentially pour gasoline into AI-based evolution.”

Building AI functions

In 2017, John Deere acquired the machine learning company Blue River Technologies. This has become one of the key parts of the company’s innovative efforts on AI and deep learning. This includes considering and constructing AI applications in machinery and other areas. “It quickly doubled or tripled the number of people working on AI,” he said. “That was the point.”

However, there is also John Deere’s data science team, with hundreds of people looking at issues. “How to analyze data from a machine and build a model to provide something of more value. Insight into producers.”

According to Sanchez, all AI initiatives at John Deere are under the umbrella of Chief Technology Officer, including organizations focused on autonomy and automation solutions. “The group is most focused on AI talent and includes the BlueRiver organization,” he added. Some organizations manage all development of the company’s digital tools (cloud, front-end mobile applications, point web solutions) with a large data science team. “I make sure they curate all the data and check all the data with the goal of producing as much insight as possible for the producer,” I explain. did.

Currently, John Deere is “quite laser-focused” on half-dozen to dozen solutions that the organization believes are most important to continue development and ultimately bring to market, Sanchez said. Said. Some of them already exist, like the new autonomous tractors.

However, the company’s goal is beyond one machine. “Our goal is to create a fully autonomous production system by 2030, which means we need an autonomous combine and sprayer and tractor planters,” he said. Today, the company offers a fully autonomous tilling solution. This is one of the four steps in the production cycle that allows farmers to prepare the land before planting. In the next eight years, Sanchez says John Deere can do it for planting, spraying and harvesting.

“Given the labor pressures in agriculture, that’s a big problem,” he said. “For decades, fewer people want to live in rural areas, which helps AI unlock.” He said this commitment to AI investment was formerly the company’s. He added that it came directly from the current CEO of John Deere, who was in charge of the big tech field. “I understood its value,” I explained.

Large-scale search for AI-driven accuracy

The agricultural industry has reached “asymptote of the value that can be added by going bigger and faster,” Sanchez continued. “The value opportunity is very important to be very accurate. You need to be able to see what you are doing, such as sowing seeds on the ground, harvesting corn kernels, spraying herbicides, etc. . “

For example, if you plant 4 or 5 corn seeds, you need to understand something about the current moisture content of the soil. Complete moisture is most likely to come out of the ground as a plant in as few days as possible. .. You will also want to analyze the quality of the soil and place the seeds where there are more nutrients. And you’ll want to make sure the seeds aren’t too close to each other, because they start competing for their nutrients. However, if they are too far apart from each other, they will not optimize the small ground for planting seeds.

“Imagine doing it on a large scale when you have to plant 100,000 acres in two weeks,” says Sanchez. “That’s why AI is already influencing agriculture. That’s why there’s a runway of opportunities. Agriculture is the best choice for AI, as opposed to wider and more generalized applications. There are all the perfect examples of. ”

John Deere’s “Holy Grail” AI Quest

John Deere continues his quest to tackle some of the big “Holy Grail” ideas about AI. One of them returns to autonomy. “To imagine a fully autonomous production system, we need to imagine an entire system that not only allows these machines to work in the field, but also understands which areas to move to next,” Sanchez said. Says. “And we need to understand how they move from field to field without a great deal of human effort.”

The second is a great opportunity for both profitability and sustainability of agriculture in that it truly understands the health of every corner of the soil used in agriculture. “That is, there is a bigger game here. It really aims to do more with less effort over time if we can farm in a way that makes the soil healthy each year. It means that you will be able to achieve it, “he explained.

However, he added, it is very difficult to measure nitrogen, potassium, sodium, etc. in real time in a reliable way. Today, someone goes out into the field, stabs a tube into the ground, takes a core sample and sends it to the lab, and after 6 weeks the results are available.

“This is like the cutting edge of current research and development. How can we measure these soil nutrient qualities in real time?” He said. “It’s really hard, no one cracked it, and there are a lot of people working on it.”

Major AI enablers haven’t come yet

Some criticize John Deere’s AI efforts, but Sanchez says whether AI-powered machines are too expensive or too complex to use, whether they own the data they collect, or whether workers change. In reality, it’s important to find good, reliable, and skilled people. Labor is one of the biggest challenges facing farmers today. Employment of agricultural workers is projected to increase by only 1% from 2019 to 2029, slower than the average for all occupations, but working on the farm is very demanding during the critical period of the year. He added that up to 18 hours of work could be required. 1 day

“Deere’s autonomous tractors and other advanced technologies allow farmers to handle short-time tasks and parts of their work, giving them the flexibility to manage emergency tasks during critical times. We are focusing on tasks that require attention, “he said. “Farmers own data and control who and when they share it.”

In any case, Sanchez claims that John Deere is still only the “second or third” of the implementation and commercialization of AI-driven solutions.

“Currently, there are three or four meaningful solutions on the market, with AI with all the sensing technology, which brings great value to hundreds and thousands of customers,” he says. increase. “But I think there are more opportunities.”

He added that “fun to think” means that the two factors that limit AI scaling are reliable training datasets and out-of-the-box computing power. Because we are using an AI solution, the more cameras and sensors we have, the more data we collect. “In other words, the more you grow, the more opportunities you have to use your dataset, a kind of network effect,” I explained.

Sanchez added that latency levels could enable John Deere to truly harness the power of cloud computing in real time, whether it’s 5G or the next level of connectivity. He added that it would take the company’s AI efforts to another level.

“So, not only are we just getting started with this, but there are some important and large enablers here that I think could make this even more exciting,” he said. ..