Breaking the phenotype bottleneck: "It can be done right here. The talent is local. And, of course, the corn is local."

Determining, analyzing, or predicting how crops will grow in the field takes time and labor. The interactions between genetics, environment and agricultural practices are challenging to measure. The newly published results of a five-year study on maize (or corn) demonstrate that autonomous ground robots can accurately and reliably capture this information.

Communications Biology has published "Breaking the field phenotyping bottleneck in maize with autonomous robots", a paper co-authored by Girish Chowdhary, a computer science and agricultural & biological engineering professor at the University of Illinois Urbana-Champaign in The Grainger College of Engineering Siebel School of Computing and Data Science  and Electrical & Computer Engineering. The team's eleven collaborators are from industry, including the Champaign-based seed and crop protection company Corteva Agriscience, the robotics firm EarthSense, and the College of Agricultural, Consumer & Environmental Sciences (ACES), Department of Crop Sciences.

The multi-year project is an example of how the Illinois tradition of cross-disciplinary research, combined with that of startups and enterprises based at the Research Park on campus, can produce results with a global impact. Chowdhary, an EarthSense co-founder, notes, "Many Grainger engineers are creating enterprises."

He says, "The strange thing is that corn is not a natural crop. What does that mean? It's not like rice, where you can go back many and many thousands of years and find ancestors. But corn, you can only find ancestry back around 8000 years, and people believe that the Mesoamericans selectively bred corn from plants like teosinte. Corn has seen the most action in terms of improvements in genetics, whether it's a traditional or a nontraditional improvement. We need to accelerate those improvements to produce more fuel and food for people."

Currently, he says, "What do breeding companies do? Companies like Corteva, or Bayer, or Syngenta? They will try to run selective breeding trials over very large areas to decide which varieties and which mixtures to selectively breed." Yield is the measurement these companies have available for use after plot trials across dozens and dozens of experimental plots. But yield is the final step in a process that includes interactive processes like genetics, environmental factors, and management practices. It would be useful to measure phenotypes or how genotypes are expressed in a crop. As Chowdhary explains, phenotypes can be physiological.

"How tall is my plant? How high is the ear of my corn? What is the width of my stem? What is the angle of my leaf? Or they could be more complex.  Is the corn resistant to this disease, for example? Today, all of this is measured by hand, by people. That's what people call the phenotyping bottleneck." Then, inevitable inaccuracies occur when multiple individuals are involved in the effort. "One person might put his caliper in one way, and another might put her caliper in a different way. One person, because of their height, might be likely to see a different part of the plant, whereas the other person sees another part. Sometimes, these could be high schoolers on a summer break. That can come with its own challenges.

Chowdhary describes the proposal in his team's paper: "We haven't automated the collection of phenotypes. That's what these robots are trying to do. We're automating the collection of phenotypes so that this phenotyping bottleneck is removed, and we can get into our computers, take our genotypes that are already sequenced, get the phenotypes, and study what they call G by E interactions. Genomics times Environment tells us which genes are responsible for which phenotypes, it's a bit more involved than that, but finding the correlations is the key. With these, we can pinpoint our breeding strategies and accelerate the pace of breeding better yielding and more resilient crops." The paper describes a long-term pilot project conducted by Corteva, a leading seed company interested in impact in terms of yield and resilience in 140 fields and 200,000 plots over five years.

Photo Credit: DISTRIBUTED AUTONOMOUS SYSTEMS LABORATORY

"For this particular study," Chowdhary says, "that was the number and it's amazing because it's not so much about replacement of labor, it's about the impossibility of gathering this data by labor. Imagine going through the fields 100-200 acres big and measuring the width of every single stem of corn! That's impossible. No human can do that, but the robots can! It takes 3 seconds for the robots to scan one experimental unit, and in those three seconds, they're measuring the stem width, the height of the plant, how many plants are in that plot, and either the ear height or the leaf area index. You have done that in three seconds, whereas if you had people do it, it would probably take them at least three hours."

Chowdhary notes, "EarthSense is a startup that I co-founded; as you know, a lot of engineering professors are doing this. I did it because I thought, 'Where are the robots?' Robots are in the lab. But if we want to take them out into the real world, they must have real-world impact. So, we had to build a team that would make those robots happen. In the lab and with PhDs it's a one-on-one thing. But when it comes to 'let's make robots that go out and scan hundreds of thousands of plots,' you need a team focused on that. That's beyond what we were able to do at the lab. There's impact, but it takes time and effort. So we had this company, and it became a commercialization partner for the DOE ARPA-E TerraMepp project. Corteva used the robot as a product. And at this point, they can continue to use it or license the technology."

The EarthSense team has moved the idea from conceptualization and testing to deployment. Chowdhary points out the difference between academic research and deployment when he says, "What we do in the university is proof-of-concept, we don't have the resources to do real-world deployment." The robot "Is, if anything, at least like a Model T, the first scaled-up automobile. It may not be the Ferrari. It may not be the ultimate machine, but it's the first scaled-up version of this. I would personally be very interested in seeing this done more because it makes sense to me, but the industry is industry. They will do whatever they do; they have ups and downs and bottom lines to meet. I believe though that good products will win in the long-term."

See a robot measure corn ear height

He says about the robots' autonomous capacity, "There's machine learning everywhere. AI is everywhere in these robots. The robots use AI to drive down the rows automatically, and then the data they collect is processed with machine learning and AI to extract the phenotype. We get images, and those images are then processed by machine learning." Michael McGuire, a U. of I. Computer Science alum (' 20), led a lot of the work on that process.

Describing the differences between federally-funded research and collaboration with private enterprise, Chowdhary says, "We are very thankful to the support we have recieved for parts of this work through DOE ARPA-E, USDA-NIFA, and NSF. Yet, we would never have been able to come to this paper without Corteva's involvement. Corteva brought to the table scale that we couldn't have created at the University level with just federal funds. We could not have engineered rugged robots and deployed them at scale without EarthSense's involvement. I think this type of research can only be done by bringing companies into the game. I think there's a very strong role that industry is playing in jointly advancing science with universities." 

He says, "I really believe in the partnership between Illinois Grainger College of Engineering and College of Agricultural and Consumer Economics, and I love the fact that the colleges have been so supportive of this type of what I would say is a little bit different way of doing research. We got computer science students who never thought they would ever see the inside of a cornfield to go and do that. And they enjoyed that; they loved it. They found purpose in that because it was for the improvement of humankind. Because agriculture is essentially the foundation of everything."

EarthSense and Corteva both have a presence in Champaign at the Research Park. Chowdhary sums it up, "The companies are local to the Midwest. This collaboration is proof that digital innovation doesn't just happen on the West Coast. It can be done right here. The talent is local. And, of course, the corn is local."

Grainger Engineering Affiliations

Girish Chowdhary is an Illinois Grainger Engineering professor of computer science with a joint appointment in the College of ACES Department of Agricultural and Biological Engineering, Center for Digital Agriculture, Coordinated Science Laboratory, and the USDA/NIFA Illinois Autonomous Farm. Girish Chowdhary is a Donald Biggar Willett Faculty Fellow.