MSU Extension MI Ag Ideas to Grow With

 - [Eric] Welcome back. For those of you who were with us this morning, we heard from several different presenters on a pretty wide range of topics. And our next one coming up is gonna be on robots. So we talked a little bit about aerial based drones or UAS. And so now we're gonna hear a little bit about ground based ones from Dr... forgive me, how do you pronounce your last name? - Chowdhary. - [Eric] Chowdhary, from the University of Illinois, he's gonna share with us about how we can utilize robots on our farms. So I'll go ahead and turn things over to you. - Yep. Thanks. Thanks Eric. I appreciate it. So, okay, so hi everyone. I'm Girish Chowdhary. I am an associate professor of Ag and bio engineering and computer science at the university of Illinois Urbana-Champaign. And it's been very exciting and interesting being here. I'm originally from India, but I did actually study aerospace engineering in Germany and then worked as aerospace engineer at the German Aerospace Center for a while, setting up their first drone program before I came to the US in 2006. And did my PhD at Georgia Tech then went to MIT for a postop and now finally I'm here in Illinois as a professor. So I have a roots in farming, although I never thought I would end up working in farming at least early on in my life. In fact, I wanted to work on airplanes and rockets, and that's kind of what my group does. We work on these autonomous robots, but the way I think, you know, I really got into this whole thing is by this idea of terraforming. And robots that can do terraforming. And initially the thinking was that, we would go to Mars and we would make the robots that would go to Mars and help terraform Mars, but the more I thought about it, it looks like we need terraforming here on earth. We're not in the best of shape in terms of where we stand with sustainability and agriculture. So a lot of the work that we've been doing in the last four years at the University of Illinois Urbana-Champaign and also at the startup that I co-founded, in the university of Illinois Technology Park, EarthSense has been focused on this idea of creating robotic solutions for transforming the earth. Through agriculture, we strongly believe that robotics and AI are the things that will make agriculture more sustainable and profitable, and kind of reverse the cycle what's going on. And of course these are big and quite ambitious goals, but they're absolutely something that I feel very strongly that more computer scientists and more engineers should be thinking about because we've all heard about how we need to increase yields before we get to 2050, right? But if you look at where we are today, and maybe we'll be able to meet the yield demands for 2050, but the cost of that in terms of sustainability metrics like greenhouse gases, loss of biodiversity, water, pollution is just overwhelming. So the question really is how can this whole cycle be brought more into balance? And how can we increase yield while also making agriculture a net positive for climate. And nowhere in the world, is this more big of a deal than in the Midwest. And particularly in Illinois, in Iowa, where we grow a lot of corn and a lot of soybean, I mean like, you know, anybody who would've come here, maybe even a hundred years ago, wouldn't have recognized the landscape. What it has become today with very large acres of corn and soybean, which is basically only sustained by lots of chemical inputs. And this is, you know, while it's good that we are able to generate this much crop, it's very difficult to believe that this can be sustained. So the question is, how do we go forward into the future and make agriculture net positive sustainable source? Overall, Ag as we know, is not a very profitable business, especially in corn and soybean where the per acre profits are a really beneath there. And expenses are keep going up. Income has been quite steady. There's another chart that I sometimes show that shows that the cost of labor has really dropped significantly over time in agriculture, because most of it is replaced by tractors, but the tractors are really replacing labor with chemicals. So the cost of inputs has really risen. And it's now even more than the cost of capital goods, which is tractors. And while we're able to convert more marginal land into use, it's been harder and harder to kind of do that at low cost. And then even this chemical heavy herbicide-based agriculture, for example, or nitrogen heavy agriculture specifically in terms of herbicides are having pretty significant issues. So we know the rise of herbicide resistant weeds all across the world, including in the US with a very significant yield reduction possibilities. Soils are significantly under stress. So large equipment are causing compaction, erosion, nutrient loss, loss of biodiversity, the reduction in organic matter. And even with all of this yield are also behind demand. So what new things can technology bring to the table? So after the green revolution, 1.0, and maybe 2.0, what do we get? What can we bring to the table that's different than what we've brought, what technology has enabled before. And if you look at it, I mean, the needle fairly falls on this whole advances in AI. So chemicals, of course improvement in genetics and reliance on chemical heavy agriculture were the key catalysts for the earlier green revolutions. So if you look forward, what can we do? What are some of the new technologies we can bring into agriculture? And I think it's really about what are some of the things that we can do better if we had more smarter equipment? And that's been a big question that my group has been asking and several other folks at Illinois, and in fact across the world are asking, and just as an example here, right? I mean, if we had under canopy robots, we could do phenotyping, which is something I'll talk about today, that will help accelerate weeding. Mechanical weeding would be an interesting topic. Reduction in nitrogen with seed and spray, under-canopy, cover crop planting, which is something also I'll talk about today. And eventually moving on to more diverse agricultural systems such as polycultures. So where are all these robots? I mean, this is not just what we are saying. I mean, people have been saying something similar to this for the last 30 years about, you know, that these robots are coming. So this, you know, the Furrow from 2007. I mean, so space age weed control is here, but where are these robots? You don't see them out there in the fields doing what they're supposed to do. And if you really look at it, right, I think the biggest problems are in this area of autonomy, I mean, I think there's a lot of hardware advances that have happened and I think to a large extent, we can say that these problems are understood, how to make robots and there's still of course challenges and I'll discuss some of them on how to bring the cost down. But the bigger, the more difficult thing that is still kind of has to be figured out is autonomy. And autonomy in Ag is hard. Part of it is just that the fields are just big and vast and scale are vast so if you just do GPS-based autonomy, I mean, that's a good start, but doesn't guarantee that you're gonna handle all the situations that existed. If you need to do something reactive in the sense that you're using data from the field to make decisions on the field, like how humans would do, then the issues that a lot of AI is built off of data sets. And these data sets from agriculture are either not available or really not good compared to greenhouse type data sets. And the robots have to deal with really difficult unstructured environments, where there is there's some structure in the sense that there are rows, but then when you actually try to get in between that row, it's kind of quite crazy in there. And then the other problem is, so not only do we need to do this, not only do we need to enable these higher levels of autonomy. Which is pretty obvious, everybody knows that, the thing is that we need to do this at a low cost. And this is the key problem that I think separates agriculture from everything else and drives innovation and agriculture, is that not only do we have to manage these high scales, but do them at low cost. So this is like a chart that shows how autonomy could progress. So each of these levels are less and less effort on the human side, but the key is that we need to do that while cost can actually come down to where chemical-based agriculture costs are today. So this whole world of autonomy, which is where kind of most of our research and technology development is focused on is kind of based off of these three different modules. And this is kind of an abstraction of how humans think about the world. So we perceive the world through our senses, the five senses that we have. And we plan some actions like, okay, I want, I need to get to the kitchen. So first I'm gonna go to this door, open that door and something like that. And then we actually execute those actions. And it's amazing, you know, like if you've seen robotics, you know that it's very difficult for us to build walking robots that can deal with all kinds of things. So as humans, as animals, we are really good at doing all these three things to a point where we really don't even notice that we do amazing amounts of competition. And we run all of this on just sugar. We don't even need GPUs or anything like that. So it's a great inspiration, and there's a long way to go, but the question is how can we get robots to at least be good enough to do specific things in a more or less autonomous way? And the big problem that happens is that you can make like the difference that is between automation and autonomy, so if you have an automated robot or automated factory thing, you assume a lot of structure, you assume that things will not change. Things will not break, but when you have a robot in the real world and is deployed at scale, it can end up in situations like this, where maybe it loses half the wing. Like this is example, kind of like a made up example in this case where we were testing the ability of the onboard software to adapt to changes like this. And we did. So this is the actual video. Let me see, okay, this is the actual, yeah. This is the actual video the robot was able to adapt to this loss of wing in mid-flight. And this is quite old now, it's about 10 years old almost. And we started on this journey of like, how do we create software that can adapt to these dynamic changes? And then do that with less and less and less input from humans. So more and more the robot is learning on its own, and also being able to capture and deal with many situations and doing real world tasks while doing this. So let me maybe take a quick break and ask if people have any questions at this point. I thought I'll continue on to the next part. I don't see anything in my Q&A. - [Eric] There's no questions in the Q&A, but everyone again, feel free to go ahead and add your questions to the Q&A as they come to you. - Yeah. Okay. So now that we've set up this idea that AI and robotics can help agriculture. The question is, how do we do this? And again, you know, we also, we set up the idea that robotics and AI can help agriculture. We also set up the caveat that it has to be at low cost. So the question is, how do we, and this is basically questions that I think I'm gonna very much answer, is how much is scalable and how can small farms benefit from this? The question is, where do you start? And as you start, how do you build up this idea of robotics for agriculture in a way that's scale effective and cost effective? So we thought a lot about this and I think the place that where we are choosing to begin is in seed breeding. And the idea is that this is where we can really get the autonomy figured out, the perception figured out. and then based off of this. We're then moving into fewer chemicals, primarily through the use of cover crops. And then eventually we wanna continue to scale this up and move into productive ecosystem. So let me give a quick idea of what we've done and achieved in there. So seed breeding is this idea of breeding, the next world idea of crop. And many of you may be surprised to know that corn, which is like everywhere here in the Midwest is not a natural crop. It's like nobody just found corn seeds lying around, but was selectively bred from teosinte by folks in South America. But we need to accelerate this breeding pipelines. And for this we need to move to more quantitative methods of agriculture, including these things like GWAS, Genome-Wide Association, where if you can figure out the genotypes correlated with specific phenotypes, then you can advance crop breeding much faster. To do this, breeders have to run these massive scale trials today. These are called, excuse me, diversity panel like trials. So here is one example where each of these grids is a three by three meter or so grid here. And each of these grids is a different variety of sorghum in this case. And the question is which one of them is doing better under these specific field conditions with the specific treatments that are happening there? This is called plot trials research. And the issue with all of this is that most of these measurements are all manual measured. So counting semi estimation stem, these are all heavily manual measurements that is kind of slowing down these breeding pipelines. So how do we automate this? So there's of course, a lot of excitement about drones, which can cover a lot of area really quickly, but they are not really good at getting some of these under canopy traits. Gantries are good, but they're limited in their area. So we really need a robot. So this is where we kind of started in 2017 and 2018. We had one of our key papers in this area where we deployed or created this 3D printed robot as a prototype that would go through the corn canopy. It was autonomous under the corn canopy, and it would use the data from its camera, just very low cost camera. So with a very big kind of this idea, a very low cost robot under, our idea was to be under $5,000 in bill of materials was actually lower than that when we made it to get important phenotypic traits such as corn count. And when we started on this journey, some of these very interesting things we can clear. Before we were building these robots, most of the robots that were built for agriculture were either really large, like the ones that John Deere makes. And they're the largest supplier of robots I would say in the world, because their tractors are more less autonomous. And in fact now there are new ones that are supposedly going to be more autonomous. And there were these types of robots that people were developing, but the issue with these robots, they were quite big for agriculture and they would often end up barely fitting in the rows and often, sometimes they would hurt the plants. And I think just before it kind of killed the plant as it was driving. So one of the ideas that we pioneered is this idea of making a much smaller robot that was able to fit between the rows of the corn. And this is an example of our very early prototype, and this was all 3D printed. And the idea was that you could drive and get good data of the farm from under the corn cap. We made 30 of these robots in 2018, and then have been making a lot of these, the EarthSense has taken this over now. And to this, I think combined with last and this year, 140 of these robots have now been deployed. And one of the things I think that engineers at EarthSense have achieved, is they've demonstrated that small isn't weak, and we can make really hard, tough field robots that are small. So let me show you some examples. So this one, this is this robot is called TerraSentia. So whenever we make robots in my group and in the places that I've helped with, each had a purpose and it's not a platform that does generic things. It's something of robot that does something very specific. This robot is designed to go under the canopy and collect data. That's all it does. It is not supposed weed, is not supposed to do anything else. It's just designed to go under the canopy and collect phenotypical data. And its purpose is to help speed of breeding pipelines. And so because of that, we made it very compact. So it's compact, so somebody can pick it up and throw it in the trunk of their car and then run off between fields. You can transport it on an airplane. You can put it in a pelican case and check it in as checked bag. It weighs only about 20 kilos with the batteries on. So you can pick it up, there's even handles underneath and it can handle most of all agricultural terrain. So that, like for example, that we are showing here now. So gravelly soils, the really grassy soils, and these are some terrain in India. So this is like a mango plantation in India, which is very dry and rocky. So, and again, like I said, you know, over 100,000 plots that been scanned with robots like and it's very compact, but still very rigid. Here's a video I wanna show you and I'm hoping that it'll make an impression because we often have this bias in our mind that small is weak. And especially in agriculture, this is very true. I mean, we believe, you know, the only things that we've seen are these humongous, ginormous, agricultural equipment. And there's a good reason for that. They have to carry a lot of payload, but it doesn't mean that small is not strong enough for agricultural tasks. So here's one of our engineers sitting on the robot and then the robot pulls a pretty heavy tractor. And it's just doing it completely on its own own power. So what I wanna showcase here is that the technology such as battery technology and motor technologies have really advanced quite a bit in the past few years, so that we can really start thinking of small robots as a practical agricultural tool that has some serious strength. - [Eric] Hey Girish. - Is there any questions? Yes. - [Eric] We've got a few questions if you wanna field some now. - Yeah. So there's some questions that I think would be better for me to answer a little bit later. So one is about monoculture, so polycultures, and then there's a question, will it dominate the market? Yes, I can answer that. And there's another question by Kyle, "You've got really plans for 10 years without any robot financially available?" Okay. So let me add, I'll answer the polyculture question later, but in terms of drones versus robots, which is something we often and get asked, I think it's, so first of all, it's not a competition because I think the robots do something very different than the drones can. The robots can see from under the plant canopy, the drones don't do that. The robots are much slower than drones in covering the field. So drones can give a very quick overview and figure out where the hotspots are, but today what's happening is that after that, somebody has to have boots on the ground. And that's, I think where the robot comes in. So we are envisioning systems where drones and robots work together. So will they dominate the market? I mean, I don't know what that even means, but if it comes to high throughput phenotyping from under the plant canopy, I think the robots are the kind of the only solution there. Then in terms of like, when are they available? I mean, the phenotyping robots are available now and they are being used by breeders and they're ready for sale. You can get them, you can talk to EarthSense. And that's something that we focused on through our ecosystem to try and commercialize this technology. And then there's a question about really deep tracks that can be nearly foot deep. So, I mean, I'm sure we can finds terrains that robots are not going to be able to traverse, but the way to solve that problem is not to go through the track, but to go around it, which is where AI comes in. So the AI is gonna be smart enough to be to avoid those types of tracks. Okay. So I think I answered some, so I'll keep going. So the next question is... - [Eric] Sorry, here's another kind of follow up question to that, so you talked about different trains. How big does the robot have to be, to be able to handle, let's say, you know, corn stove or if you've got a no-till system? All the videos they kinda looked flat. - Yeah. I think that that's not an issue like these robots, the ones that we have right now, they can go over quite a bit of stubble on the ground. And then I'll show you another robot later on. That's a bigger robot that's designed more for operation in a field. So the really small robot again, I think that's the key thing. Like if I can get one message through today. As more of an educational message , is that, let's stop thinking of robots as replacement for humans, okay? They are tools that do specific jobs. So this robot is not going to be a harvesting robot. It's not going to be a weeding robot. This particular robot is designed to do one thing, which is go through the field and collect high throughput phenotypic data. And it operates in fields that are more or less designed for high throughput phenotyping. Now I'll show you another robot that's designed to go through the fields and plant cover crops. And that is designed to operate in more production fields. And then we can make other robots that can go into no-till systems or in places where there are foot long ditches, but this general, I think what is slowing down robotics in a large way is that people try to make or try or expect these amazing beings that will do everything in the field. And that's just 50 years away. But what we can do today is automate specific tasks in a more holistic manner, which is what I'm calling autonomy. So I'm hoping that makes sense. "So are the robots available in Michigan?" Absolutely. They're available in Michigan. If you would like to get one, you can. I think there's already one in Michigan State University. Yes. Addie Thompson has one at Michigan state. Okay. So the one thing that's different with these robots is that you cannot rely on GPS to make them fully autonomous, just like bigger tractors are, because GPS is not always visible when you go under the canopy. So even with RTK GPS, which would get you a few centimeters accuracy outside of the canopy, when you go inside of the canopy that accuracy drops. So for this purpose, and I'm gonna skip some of the technology stuff, but just to give you an idea, some of the key advances we have made, is in creating autonomy systems that are reactive to the plants. So we have, for example, in our beginning work, we were using the LIDAR, which is a laser device that determines the distance and angle with respect to the plants. And just to give you kind of a flavor for those of you who are into technology. The problem that we are solving is that, this is the point cloud or the LIDAR is basically like a spinning laser radar type system, which gives us distance to a bunch of things that the robot sees. So the robot is here, is going forward. And this is what we are seeing. The trick or the algorithm that we have to design the AI has to figure out where those rows are with respect, you know, with this crazy point cloud that it's getting and doing that of course is technologically challenging. You know, and it's describing one of these papers. We have more papers coming now out on this idea, of using LIDAR for road based navigation. You know, we've been designing these, and one of the things that we insist on in my group and overall with the Center for Digital Ag and AI farms, at Illinois Urbana-Champaign line is that the outputs of our work live beyond papers. So you can always write a, you know, you can make things that work for a few meters and then get published. But we really insist on things that are robust and are reliable in field environment environment. So everything that we do, all our autonomy algorithms go through this kick test, where basically either me or somebody else is kicking the autonomy, the robot to see if it's working. So the main point being these autonomy methods are getting more and more reliable. So with these, so here's the product that has now actually made an impact in agriculture. So this is the TerraSentia robot. It goes through the canopy and it gets data from its cameras. And using that data, the phenotyping system is able to deliver traits of value to the breeder, including things like stem width, corn near height, plant height, corn count, soybean pod count and a host of other traits that are of value. And these traits are collected by the robot. They're processed online on a cloud, and EarthSense has put together like this whole system for moving all of this data. And at the end of the day the breeders, these large companies that work with these types of robots get like a field map as to how each of their plots are doing. So that's pretty exciting, that some of these technologies making it out there and robots are actually coming out now into the fields. And I'll also mention that, there's also lots of other success stories today in agriculture with robotics. And it's really getting pretty exciting. So Naio is here, they're delivering weeding robots. They're not designed for corn yet, corn and soybean yet, but I'm sure eventually there will be weeding robots for corn and soybeans also. Which brings us into the next part. Let me see if there's any other questions. So, are there any questions on the phenotyping part at the top? - [Eric] I think these are all the same ones you saw before. So I guess we'll handle those later. - So, I'll move on now to more production facing agriculture. So what do we do in production agriculture? So, which is today heavily chemical-based. As we talked about in the beginning, and there's issues such as herbicide resistance on this side of the planet, and then on the other side of the planet, there's issues such as labor, mismanagement or basically in, I don't know, you know, just a big mismatch between the work that is being done and what kind of labor is available and what that labor can be paid. So the fields, for example, in India are much smaller and which relates back to the small farm question that we had before. And these small farmers are not able to afford larger equipment, which would help them automate their fields. So in India and other case places in Asia, often people rely on human labor. Now this human labor is not easily available and often would prefer doing other jobs than this because this doesn't pay as much and there's better jobs that they can do in the cities. So there's a pretty big fix that most of these farms are in, in terms of what they can do to help automate some of these things and it has to be small robots that I think would be a good solution if we can get them to work there. Again, like I said before, I mean, if you look at even US, while the cost of labor is dropping, it's pretty clear that the cost of inputs is rising significantly. So it's the money that farmers are keeping is consistently reducing. And this has led to basically these large scale monocultures and monocultures that are like so superiorly incentivized on per acre costs, that we are often forgetting about some of the sustainability things. And even when some of these options are available, farmers are reluctant to try them. Like no-till is one, you know, organic is another, but even cover crops which everybody could do in theory. And apparently have benefits that are proven at least in trials, but maybe not at large scale because nobody's trying them at large scale or some of these issues. So as a result a lot of the land is just bare across the year. And I got an example... (strong wind whooshing) I recorded this video the other day after Thanksgiving flying into Champaign, and you just see like this, this is what the corn and soybean desert. I mean after the fall, a harvest is just brown land. So one solution could be to have a robotic system, let's say several issues with cover crops, right? So people don't wanna try cover crops because it costs quite a bit per acre, $15-$20. And at least in the more Northern latitudes, we can only get it done after the harvest because there isn't a good solution to do it at good cost during the season. So one idea would be, you make a robot that is bigger, more sturdier, supposed to go through the rows of corn and plant cover crop. So this is our cover crop robot, which we're code naming TerraPreta right now. I can take 120 pounds of seeds and working in teams. So a team of five robots, working fully autonomously, the idea is they can cover 80 acres of a field in about five hours. And this is the kicker. So if we do it with this way, it looks pretty convincingly that it should be able to be done at less than $5 an acre. So not only can this robotic technologies enable options that are more sustainable for management, but they can do it at a fraction of the cost than what is possible with larger equipment. And so how is this cost reduction possible? So one part of it is just the cost of making these robots. It's really not that much because, it's just a bunch of, it's some metal here, but compared to like a large tractor, the operation is really simple making this. And then the other part of it is that they're more less autonomous, right? So the labor cost is really removed from the equation. And, you know, in order to do this, we've been solving this problem of full field autonomy. So you go through the row, you turn, you come back, you go back into the row again, and this is harder than it sounds because in a real, kind of a farm setting, all kinds of issues exist, right? So for example, part of the row may be missing. Then GPS doesn't work well inside, but when you go outside, you have to use GPS in order to go to the next row, right? Because otherwise you have to count rows, which is a whole different issue. So we've been working on these technologies and papers are coming out now about how some of that is done. And also this has been integrated into some of these products that we have, but here's an example of the types of things that robots have to be able to do. So if you have a wet day and the robot's out there, the wheels are all gonna get cake with mud. So the types of algorithms that you have need to be able to adapt to these difficult kind of changed situations. So in this case, the robot process to adapt, like making a turn with heavier wheels automatically, and then it actually makes a mistake as it's trying to go in. And so automatically has to detect that it has made a mistake and then try again the same maneuver. So it backs up in this case and then tries again. So these behavior is making them autonomous is interesting and exciting. And over time, the distance between interventions, which is a key metric that we use has been increasing. So what I'm telling you is that in a few years, or maybe not even a year, maybe in a year or so, we'll probably be able to do very long paths autonomously through these fields. In fact, that's been deployed on some of the robots as we speak. Now, the other thing we're doing is also bringing the cost down. And one way we're doing that is we are getting rid of the LIDAR. So the LIDAR is, like I said, is a laser based device, which has to be expensive because it has to have a good clock on it because it measures the difference between time of flight of laser light. So instead, one way we could do this is using cameras. So cameras that are on the robot cameras are much cheaper today to be able to go through these fields. But it's very challenging task for machine vision, because there's a lot of issues in out of clutter. There's a lot of green on green in the fields. So how do we do that? We've been working on this machine vision algorithm, which I'm kind of skipping the details, but just to give you an idea, there's a camera in front of the robot. It's getting every frame. And then based on the frame is deciding what the angle of the robot is respect to the row and how far is it from on the row. And then using that and optimizing those neural networks that do this, we're able to not only reliably drive further, but also drive faster than we'd be able to do with LIDAR, because we can better differentiate between geometric obstacles and obstacles that are geometrically, not traversable, but physically the robot could go through them. Whereas the LIDAR would've hard plan differentiating between those two. And again, you know, these methods are fairly robustified now, and this is again, one of my students, Aaron, testing the robot with the kick test. Also the algorithms are getting better. You can see that this robot is slowing down when Aaron really kicks it hard, which means it's kind of aware of its own speed and that it knows that in order for it to recover, it has to slow down, which is as a human it's like a second nature behavior for us, but it's cool that robots are learning to do that. Another thing that happens is that it switches seamlessly from soybean to corn at some point. So this is early season soybean. This is the autonomous farm that we have at Illinois. So Aaron gives up, I think he has had enough soccer practice, but then eventually the robot goes from a corn field just directly into this, sorry from a soybean field into a corn field and continues to be autonomous. And then we continued testing it. So this is now, again, this is a video I received from EarthSense the other day, they've been deploying some of this. So this is an example of what they're doing. So the robot goes into the field. These are breeding plots. So it stops, verifies all sensors are good, scans the plot by driving through it autonomously, able to handle inconsistencies in the plot, then comes out at the other end, stops again, verifies the data and then makes a maneuver to get into the next row. And now this is again, ready for deployment at a pretty large scale across corn fields. And then it again, you know, reads and repeat. So this is very exciting. I think this is a new age now in how we can interact with fields, get data. And of course eventually the idea is that these robots will work in teams to cover larger fields. So here's an example of how we've been doing cover crops with these. So you have a robot, a team of robots that have the seeds in the seed bin and the robot drives through the field of autonomously. And there's a seed dispenser at the back. And there's many different designs that we we're trying. And then you broadcast this fields really close to the ground. The seeds really close the ground in mid-season. So we can start doing this in June and July. So even when, you know, before harvest, there's a pretty good stand of cover crop, and this can be done at a much lower cost. So we've done 100 acres last year. And if anybody's interested in partnering with us this year, we're targeting to do a few thousand acres this year. So it would be interesting to work on that. Okay, so now let me answer the final kind of bit, is how do we move towards polycultures? And there was a question about that before. So, like the trick to getting to polycultures, to a more diversified agriculture is this idea of plant manipulation. So here is a great example of polyculture that I like to show, it's an oasis, right? So in the Indian cultures, these oasis have been always a great example of how to use minimal resources to create maximum production while really nurturing the ecosystem, because it's a very delicate ecosystem. So there's plantations at different layers and they're taken care of very much. And the only way this is possible is with a lot of dextrous human labor, right? Another exciting example that I think really has a lot is urban gardens and whether they're rooftop or not doesn't matter, but urban gardens that take small areas to produce more food for the local areas can really help offset some of these food disparities that we have. So to get there we need to make robots that are dextrous and can scale up as opposed to tractors and tractor-based heavy equipment that is not very dextrous, but can scale up and humans that are very dextrous, but, you know, manual labor doesn't scale up. So how do we do that? So this is more research that's happening at the university. I think this is five years, if not more away from production, but we are really excited about how things are going. And the idea is making these soft robots. Robots that are made of, similar to octopus arms, or elephant trunks that can bend around and maneuver around plants to do plant manipulation. And here's one example of these robots where there's a regular rigid arm, and then this soft robot comes out of it. And the idea is that, if you don't need to extend the soft arm, you don't do it. But if you do, like in this case, when the berry is inside, you extend that arm and you grab that berry. And this is a very active area of research for us, where we are trying to improve how the robots can navigate using cameras on board, the robot are also on the arm and reach and get more and more complex berries and things. Okay, so I'll kind of end right now and open up the floor for questions, but just a big thank you to the National AI Institute that we have the AIFarms, which involves a number of different institutions and the Illinois Autonomous Farm, which is one of the key places where we'll be seeing all these examples. I'll finally end with this fun video of the farm bot that we have over there. And this farm bot is a system that you can buy of the shelf. It's a CNC type robot. And the idea is to kind of create like a little test bed where you can try and practice fully automated gardening, And the idea is that if these types of gardens can be made, then you know, they could supplement income in communities. And of course, I mean the idea is pretty good. There's a lot of AI work that is going on in, because right now it needs like pretty precise and somebody is to really program it well for it to be able to do this task. What we are trying to do is simplify the programming. Part of that is talking to the robot through a national language. Other part of that is helping the robot understand what plants there are, so instead of saying, go to location X, Y, Z, and do action C you know, it'd be nice if the farmer could say, or in the homeowner could say to the robot, "Pick tomatoes that are ripe." And I think the plant that I planted first in this season is probably more ripe than the others. So if robots can deal with that type of information, then we are really moving towards robotic solutions for farming. So that's it. I'll take more questions. So what was the question? We have been planting for robot ready plans for 10 years, without any robots financial available, these robotic system seem to be more susceptible to pessimism. So Kyle, can you explain what these robot ready systems you're talking about? - [Eric] Yep. While Kyle is typing that in, do you wanna handle that first question of scalability? - Yes, absolutely. That's the whole idea. And this, the idea is small robots for small farms, and then the technology is scale neutral in the sense that if you have larger farms, we can get more robots, but so they work in teams, but if you have a smaller farm and you only need one or two robots, and you can do that, this is different than larger sprayers and bigger equipment which you need to have a significant operation to be able to even afford that. So that's absolutely the idea. Yeah. So apple orchards. Yeah. Apple orchards, I don't know much about apple orchards honestly, to be able to answer that question properly. I do know that there's a lot of work going on in automating apple orchards and robots based systems. I think that you're absolutely right. The issue is that robots are not there. And there are a couple of companies that were trying to do it, but they've gone under at this point. Although there is this one company that is doing spraying in orchards, pretty good in Almond orchards, I forgot their name. Their robot looks like a giant rocket ship, but they're doing pretty good with the spraying. So maybe, you know, those robots are coming to apples. Any other questions? I'd answer a question that everybody asks me, what is the battery life of these robots? So on the TerraSentia robots, we get about four hours. And then on the TerraPreta, which at the cover robots, we've had about five, six hours. We haven't done a big evaluation yet, but we're pretty comfortable saying that the battery life can be much higher than, it's definitely, obviously higher than a drone, which is what most people are biased by because drones just, you know, are very battery hungry because they have to keep themselves up in the air. - [Eric] So if you were to have a, let's say a small fleet of these, and you're in a somewhat remote location, you're not right near the farmstead. Do you have, or do you envision, battery charging stations that you would then take out, they would dock in the field then go back out to... What systems do you have in place for that? - Yeah, absolutely. Yeah. I mean, I think that would be one way. So there's a couple of ways they can be deployed, right? So one way would be that there is a traveling set of robots. So there's a robot team. So let's say each robot team has maybe say five robots and then farmers sign up and say, "I want this 80 acres covered today." And then the company, or whoever is supplying the robot as a service shows up, drops the robots. The robots get the job done. And then they're picked up and moved on to the next field and they're washed and cleaned before that happens. That's one approach. And the other approach is kind of this permanent robot idea. So that you have this field and there's a group of robots just dedicated to that field. And whenever they need a battery charge, they come back to the battery station and they get a new battery. They get a cleanup in this recharge station, maintenance farm and then they go back out again into the field. So both options I think are definitely something to look forward to. - [Eric] OK. There's more... (Girish interrupting) - Sorry, I was just gonna say robots at this service will probably be the near term option. Go ahead. - [Eric] So there's no more questions in the Q&A, so folks, we still got about 10 minutes, feel free to drop your questions in there. - Yeah. Just ask anything that you think that you'd be interested in knowing about robots, whatever you want. - [Eric] So, one thing that I was noticing is that you were showing all these different videos is they're all going down the row, have you worked with these robots in headlands? You know, so if you've got now a corn plant right in the middle of the row, how do they handle the headlands? - Yeah. So in these videos, they were kind of focused on that, but so basically the headlands turn maneuvers on the ones that I was showing you, are assuming that they're pretty clear, but we have other work that I haven't shown here in which we trying to solve the traversibility problem, which is the robot uses the cameras onboard the robot to decide how to change its path. If there's like obstacles in the way such as corn plants. And then the other thing, I mean, a single corn plant in the row in the headline is not a big deal for the robot it just goes over, but if you have these, you know, like some people do end rows, right? So you have some rows, let's say you're planting mostly East-West, but the end rows are North-South. So in this situation, there is no gap for the under canopy robot to come out, so that would've to change. So there would have to be maybe at least one skipped row there, so that the robots could maneuver through there. - [Eric] So, you talked mainly about, you know, camera sensors, helping with breeding, things like that. You kind of mentioned having other tasks for the robots to do like weeding or whatnot. Here's a question about maybe attaching a mower to it, or what can you envision some other tasks, these small robots doing? - Yeah. So weeding is absolutely a big thing. I didn't talk about it much. Mechanical weeding is, I'm sure is becoming a bigger and bigger deal in Missouri, certainly is a big deal here in Illinois now with all the herbicide resistance. So that's absolutely something that is in the offing, even with the work that we are doing, we've done a bunch of simulation studies on it, which I didn't talk about it in this talk, but my PhD students, he established that it's possible to, it's feasible actually for these robots to actually go through and, and make sure that fields are weeded. So that's there, and then we've tried a bunch of hardware on the robots, and I think it's promising to some extent, but we're trying to solve the cover crop problem first, just to get the scale in. And then I guess the next thing we would do is attach mowers or whatever people are saying, some kind of devices on the robots to scale up the mechanical weeding. - [Eric] Next question, operating tractors, combines, field sprayers in the future? - So I don't know what that means, is the question about autonomy of tractors or combines? - [Eric] Yeah. Good question. Iowa, could you maybe just clarify a little bit what you mean? We'll go to the next question. Anyway these can be subsidized so to encourage the sustainability of agriculture? - Absolutely. Absolutely. Absolutely. Yes. I think cover cropping would be a great start. So adding, cover crops have a number of sustainability benefits and at the cost that we are talking about $5. And I think there are some simulations that I don't even talk about. I mean, I believe them, that I know that they're true, but the numbers just sound unbelievable, is that we could do it at a dollar an acre, right? So at that point, I think it really begs the question where all these companies that are claiming to increase or decrease their carbon footprint by sequestering carbon could just take on the cost of planting cover crops at a dollar an acre. And the farmers can enjoy the benefits because their land is improving and the companies can sequester their carbon. So that's one way, another way would be to work with legislators, to see if cover crops can be subsidized, these under canopy methods, because there's so much lower cost than the other methods could be a good place to start. Of course they need to be de-risked, but yeah, and mechanical weeding would be another. I mean, you has moved to say that 50% chemical reduction has to happen in the near future. So I think those types of restrictions are coming, I hope also incentives come. - [Eric] So the follow up clarification there, talking about robots operating farm equipment. So we've got an existing piece of equipment, say a tractor, a combine field sprayer. And I think the clarification is, "Do you envision robots operating existing pieces of equipment?" - Yeah. I mean, I think what we would do is take the brain from the robot and put it in that equipment. That would be like a very rough way of saying how that would be done because I mean it doesn't make sense for the little robot to pull a big robot or something. So, but yes, I think, and absolutely that's the idea is to try and make more and more autonomous farm equipment and also bring the size and cost down so that it's actually applicable to smaller farms as well as bigger farms. - [Eric] So we may not have time to address all of these, but Reuben has a couple of different questions in here that are actually a few questions within questions. So you can just probably read through those and answer them as you go. - Yeah. I think the one about nucleic acid extraction, I don't even know how to answer quickly. We have not thought about nucleic acid extraction. That's how I'm gonna answer that. And then the second question, that's a good thought though. We will definitely look into that. So the DNA recovery also, I don't know. I think what Reuben is maybe trying to get at, is that there are these really exciting possibilities. Once we are able to manipulate plants in the field at an individual plant level, and maybe that's where some of these newer really state of the art or beyond state of the art technologies could be brought to bear. How do we keep robots from making people irrelevant? I think that's a very good question. So the issue with agriculture today is that, well, first of all, robots are not replacement of people, that's why I've been trying to say right from the beginning, I think there's a big fallacy in robotics. What we call the substitution fallacy, where people often think that robots would just, you know, replace a job. But the reality is that that's not what happens in the real world. A lot of things have to change for a robot to be able to do its job. And the whole system kind of really changes around this. So if you look at agriculture today, the issue is that labor is difficult to find and then labor would prefer doing something else rather than work in the fields. They're probably the lowest paid jobs. It's not like people are lining up to go and do these jobs. So I think that's a very key factor why even robots are applicable today. And if you think about things like mechanical weeding. I mean, of course, you know, people could go and do that, but that's not going to be sustainable in terms of what the farmers would do. So the question really boils down to, do we enable farmers to be profitable by helping them be more efficient or not? And I think if farmers can be more profitable by being able to utilize their land better, they will have more income sources. And that will improve the economic conditions in rural areas. So I think that's the solution, but I don't think we're having people free farms anytime soon, although that's a great academic goal to have.