When Will Autonomous Farm Equipment Actually Pay? | AgCast #206

Audio Transcript:

Chad Fiechter: Welcome to the Commercial AgCast. I guess I’m your host by default. Chad Fiechter, assistant professor in the Department of Agricultural Economics. I’m also the research director of the Center for Commercial Agriculture. I’ve invited, and I would call you a co-host, not a guest, because I think you have a very impactful role to play. And this is Josh Stine.

We are going to talk about large scale autonomous machines.

We wanted to understand the economics. There’s a world where an autonomous machine could be more efficient than human operation. Most of the multinational farm machinery companies announced that they would make the corn and soybean production cycle autonomous by 2030.

Josh, do you wanna say who you are and why you’re here?

Josh Strine: Hello, I’m Josh I’m currently a third year PhD candidate here at Purdue. Excited to talk about some research that I’ve done with Chad.

Chad Fiechter: Perfect. The reason that Josh is on this podcast is Josh and I, along with, James Lowenberg-DeBoer, who our listener may know, wrote a paper recently on, the economics of large scale autonomous farm machinery.

Josh Strine: We’re looking at the large farm equipment that you see someone sitting in that tractor. But there’s a lot of conversation going on that going forward that may not be the case. So we’re thinking about these large, tractors that currently run or usually used, but being operated autonomous. So just having no one in those cabs.

And we specify large scale. ’cause a lot of research is out there that focuses on autonomy, but a lot more discussion about smaller robots that you see that usually don’t even have the size to fit a person in the cab.

Chad Fiechter: Yeah, perfect. So, and, and the reason, Jess is a key part of this collaboration is that Jess has done a ton of work on the economics of autonomous machines. Primarily those small scale swarm robots, like, drone spraying swarm robots, weeding robots. He actually operates, at the university, an entire project where they raise cereal grains, I think it’s wheat and canola with autonomous machines. But on a very small scale, I think their largest tractor might be 40 horsepower.

Okay. So, just to set the stage, the reason we are interested is I think most of the multinational farm machinery companies announced that they would make the corn and soybean rotation cycle, the full production cycle autonomous by 2030. So we wanted to understand the economics. So we use something that is a Purdue legacy to do this work.

Josh Strine: Yeah, so we started with the PCLP model, which I believe stands for the Purdue Crop Linear Program. It was used as an extension tool, so it was created for this way for farmers to come in and, kind of plug in their own farm and parameters and see what kind of decisions they can make to optimize production. And really optimization is the key word there with what we’re looking at.

So we took this tool and we created our own case study farm that we thought maybe represent, an Indiana farmer, a Midwest farm. So really we went in and we optimized this case study farm, but we did it under a bunch of different scenarios to compare what we conventionally see versus some of these autonomy potential outcomes.

Chad Fiechter: I wanna go back ’cause I just, I was at an extension event yesterday where someone told me about running this model very, very long ago. Where you would used to take these cards that you would program and you would take ’em and run ’em in a mainframe of a computer and it would take a significant amount of time running on a mainframe for each individual solution.

What you did on your personal laptop is you ran solutions for how many optimized farms?

Josh Strine: So since we were unsure about autonomy in a range of different ways, I wanted to run all of the variations of possible outcomes. And in order to do that, we ran 72,000 simulations. It didn’t take days necessarily to run one simulation, but to run through each of those possible farms, I think in total it took about three or four days of time running on my computer.

Chad Fiechter: Yeah. That’s amazing.

Chad Fiechter: Okay, solet’s get into the dimensions of large scale autonomous machines that we defined as important to understand the economic performance.

Josh Strine: The variation and uncertainty around autonomy is what led to these 72,000 simulations. And so when we’re thinking about autonomy, how it interacts with our profitability to farm, we came down with like five or six key different levels of variation.

One of those is the autonomy intervention time. So we don’t know how much time it’s gonna require for an individual to either put eyes on the equipment or look at the technology that they may have in the office to keep track of that equipment that is running.

Additionally we have the efficiency. So compared to a human sitting in that cab, how fast is it able to manage the operations in the field?

We have the hardware costs, so this is, thought of as an upfront cost of the technology that they physically put into the equipment.

And then also related to the autonomy as maybe a subscription fee. So. When we first did this, we thought about a per acre fee that may be required, each time it’s used. So if you think of a annual production plan that may have a passive tillage, a seeding pass, and then maybe spraying or harvest, each time you make one of those passes would have an additional fee. That was an assumption that we made.

And then outside of the autonomy, there’s some other things that kind of interplay with the profitability and relative profitability related to labor. So when we think about this autonomous, it may be more profitable if labor becomes more expensive. So we have to look at that change in labor rates and then also just labor availability. As we have less labor available, then you can’t do as much yourself, so then you’re gonna need that additional autonomy equipment, or it may be better suited in that circumstance.

Chad Fiechter: All of those parameters was with conversations with farmers, people who have interacted with these machines so far. And we came down to there’s the current performance of what we observe, at least demonstrated today, and we had to make some guesses. Right.

Josh Strine: Yeah. So, based on what we thought, we, we worked with the autonomous intervention time. We went with 10%. So that sets out to be six minutes of every hour that the machine runs.

For the efficiency. We went with 80% of a human.

I don’t remember what our autonomy hardware cost was. I think it might have been 40,000, but it’s secured there in the paper

Chad Fiechter: Well hang on. The 40,000 is the cameras, the valves. It’s to make a machine autonomous.

Josh Strine: Yes, that is to the upfront technology cost. We did 40,000 for the unit.

Chad Fiechter: Yeah. Each power unit.

Josh Strine: Each power unit. Right? Each power unit. Our subscription fee.

We did it first based on a per acre, but if you look at the paper, the number’s a little weird because we had to transition that to hector. I think $3 per acre.

Chad Fiechter: Which corresponds, I believe, to a $7 and 41 cents per hector.

Josh Strine: Yeah, that would, that would be right. We worked with $30 per hour of hired labor for that rate, just because that compares the most with our autonomy outcomes.

And then one thing I did forget to mention earlier, which may be a key player here, is the machinery hours. So in this autonomy, we may allow those machines to run longer per day because you don’t necessarily need a person sitting in the cab. And so in the, current. Expectation, we thought maybe 18 hours of tractor, time per day.

Chad Fiechter: Okay. So now we’re gonna talk through our results. ’cause it’s fairly quick. And one of the reasons I wanted to have this conversation with you today, Josh, is because we’ve been quoted in a couple of popular farm outlets. And we are dissatisfied. We need to be more direct in how we communicate this.

Chad Fiechter: So I’m gonna jump to our first set of results where what we’re comparing is a farm of similar size operated, where the machines are conventionally operated by humans, and one where all of the machines are operated autonomously. So the same tractor in one scenario is driven by a human, and that same tractor in the autonomous scenario is driven by machine.

Josh Strine: We know this model isn’t a hundred percent right, so we obviously aren’t gonna sit here and say how much profit any of these farms will make.

Chad Fiechter: We can both come up with reasons why it’s wrong.

Josh Strine: Yes. What we really wanna focus on here is the relative profitability of these different scenarios.

And so we compared everything to a baseline, which was just a conventional farm, a fixed farm size, all of it is owner or hired labor. And then comparing that against some autonomy scenarios, whether labor is hired or labor is not hired. And what we just described is maybe the realistic short term or a more ideal long term.

Chad Fiechter: And just for clarity, that ideal is like when autonomous machines operate the same as a human, you can run ’em just as long as you could if there was somebody in it. We have minimal, costs for subscription fees. Right? That’s kind of how we’re thinking about ideal.

Josh Strine: Exactly. A hundred percent efficiency, minimum oversight, longer time running per day. All of the parameters were set to the maximum in our range basically.

So looking at the results as we compare all of those to our baseline, when labor is available, we don’t see in either the ideal or the current realistic, situation that the autonomy is more profitable. We see looking at the numbers right around three bucks an acre.

In the ideal,

Chad Fiechter: If autonomy reaches perfection.

Josh Strine: Yes, and maybe our more realistic short term case, we see a deficit of $63 per hector. Divide that by 2.5 and you’ll get your per acre.

Chad Fiechter: Okay. So I wanna go to the part that both of us have been a little bit frustrated with. We, we have not done a good job communicating, so, so I’m gonna give us each a chance to communicate correctly.

What would take away that $7 and 70 cents per hectare decrease. What would the labor rate have to be for you to be better off? If we increased how much we have to pay people for labor where you would still be better off with autonomous machines?

Yes.

In the ideal case, it’s 45. So if autonomy reaches what we consider, the ideal scenario and the labor rates were $45 an hour, which there’s probably some farms that, that it with all of the indirect cost of having an employee could be getting close,

Josh Strine: Yeah.

Chad Fiechter: like we could make the argument. Right. $45 an hour.

But now if we go to the, the most likely scenario, what we observe, large scale, autonomous machines doing today, what’s the labor rate that we would need to haveto break even?

Josh Strine: Yeah, when we look out there, right now, we would expect labor rates to have to be around $140 for autonomy to break even with our

Chad Fiechter: Now you are also an expert in ag jobs. Is, is that fair?

Josh Strine: Let’s say I dabbled.

Chad Fiechter: Okay. You dabbled in ag jobs, how many of those ag jobs were $140 an hour labor rates?

Josh Strine: Not very many.I don’t know if any of ’em were uh, farm machinery operators.

Chad Fiechter: So I think this is the part where we’ve struggled to communicate very directly is that we’re not in the ballpark, currently. The current economics around this are not in the ballpark.

Chad Fiechter: But there is potential that we have yet to capture the actual efficiency that could be gained. So there’s a world, like engineers tell us, there’s a world where an autonomous machine could be more efficient than human operation, which would vastly change our numbers.

Josh Strine: Yes, a hundred percent.

Chad Fiechter: Right, but that’s probably not a three to five year sort of thing. That’s probably a ways out.

Josh Strine: No, it’s not there yet.

Chad Fiechter: An interesting thing that we didn’t include in this model. So we, we’ve talked about. Well, we just explained our results of, let’s assume labor is unlimited. We did look at the alternative, I guess the opposite end of the spectrum where there’s no labor available. And in that case, when we actually start running into this labor constraint, very quickly, autonomy makes sense.

Josh Strine: Because if you don’t have autonomy, the owner’s labor in themselves, is not enough to farm the whole farm. So it just makes sense to get that technology to be able to operate all the land. After having some of these conversations since we published this paper, it got some questions about, well, what if we had a partial labor constraint, and so that would be something interesting to explore. We didn’t necessarily do it here because we’ve already said we had 72,000 simulations and we

Chad Fiechter: Need to call it somewhere.

Josh Strine: Yeah. Uh, so every multiple of that labor constraint we add just multiplies our number of simulations. So now that we have an ideal in a current scenario, we can look at that labor constraint and see how it may affect profitability.

Josh Strine: Another big one is looking at using autonomy for specific, production practices. So maybe there are, things out there, part of the production plan that would be more ideally used for autonomy. Maybe you use it for tillage, but nothing else. Maybe you use it during harvest when you already have another piece of machinery in the field, so you don’t necessarily need to oversee it. You’re still there. So exploring that specialized use of autonomy towards a specific production practice may be of interest.

Chad Fiechter: Jess was one of the first to point out, maybe we don’t really understand why we adopt technology. And he pointed out the case of milking robots and it being life improving. Can you give us your thoughts on that? Is it gonna make people’s lives better to the tune of sort of whatever it would cost?

Josh Strine: Yeah. I’m gonna say that’s gonna depend probably on personal opinions. If we think about some of the stories I’ve read, you can now have the machines running in the field, and be at home with your family, watching a kid play basketball or going, going out to a sport.

So it certainly provides an opportunity to be spending your time elsewhere, maybe doing something enjoyable. Or maybe it’s doing other work around the farm, whether it’s paperwork or if it’s doing something back at the shop to get ahead of the next step. So there’s certainly opportunities that the second you don’t have a person in the cab, you’re doing something else. Whether that’s worth the cost of this technology is probably gonna vary by each person’s preference, but there, there are opportunities there.

Chad Fiechter: Alright, so what do you say to those people who say, Josh, you just hate technology? You just don’t wanna see it succeed.

Josh Strine: I would say that I don’t hate technology. I use technology every day in my life, in my work.

Chad Fiechter: That is like the least way to characterize Josh as he sits here with his iPad, his, his MacBook. There’s a litany of technology that surrounds you. This is outta curiosity, right? Like this whole project didn’t start from trying to shed negative light on the possibility. It was just to try to understand where are we at currently? And what do you think is some of the first places that you think there will be a chance for autonomous machines to be involved in sort of Midwestern agriculture?

Josh Strine: Based on if there is labor constraint, obviously that is number one.

Chad Fiechter: Meaning it does not exist. You cannot hire anyone.

Josh Strine: Yes. So that’s very quickly the number one place that it could come in.

Chad Fiechter: Just for one other caveat, you can’t hire them for $145 an hour. You literally can’t find someone who that will work for 145. Like just to be incredibly precise.

Josh Strine: Yeah. So the labor is not available at that steep rate.

Chad Fiechter: When I was a producer, I also said sometimes that I couldn’t hire anyone. And it maybe was more reflective of how much I was willing to pay that person.

Josh Strine: Yeah, that’s fair.

So obviously the constraint labor. The second opportunity, may be farm expansion or larger sized farms. So although we’re assuming this kind of fixed subscription fee on every acre farmed, by expanding the number of hours that this tractor may be running in the field a day. You can still get more production done, within that window of possibility in terms of planting and harvesting. So that may be an opportunity that with the same set of equipment we see it can farm more acres, than it can conventional. Just based on running more hours a day.

Chad Fiechter: Yeah. And I also wonder, a couple of the dimensions we didn’t explore is individual practices, maybe tillage, like I think that’s the one that I’ve heard the most. Tillage may be one of the easiest places to automate, for the simplicity of the operation. Then thinking about the idea of maybe you have paired, right, you have one, a human operating one and a paired, solution, running in tandem too. And that’s something we didn’t really cover.

Is there anything else that you are interested in taking this thing a little bit further?

Josh Strine: I think those are the places I would start. Looking at that labor constraint and finding where that binding percent of labor is. So when we didn’t include a constraint of labor, I think the most we ever had to “higher in a month” was less than 800 hours. I don’t remember what it was exactly, but when looking at the other research, that 800 hours was usually what they set the constraint at. So we’re already below what other research has said is maybe constraining labor. So finding that actual spot that it binds, looking at some profitability, comparing these individual practices, or even building the choice of autonomy into the model.

So the way we did it here, was we assumed that the farmers either all autonomy or all conventional. And building in that model a choice for autonomy, like maybe we can then, instead of self diagnosing when they may choose it, we can let this optimizer choose where autonomy would make sense. So I think this, as far as we know, is the first to really look at these large scale autonomy. So there’s a lot of room to grow and a lot more things to figure out.

Chad Fiechter: I think it brought up in both of our minds as we were working on this project, some indirect or some related questions around how is autonomy gonna impact Midwestern row crops? The first one for me was, are we gonna grow different things? Right? So we shifted to maximizing one person’s time so drastically, that, you know, we’ve got row crops grown here over and over again. And will it be that autonomy would allow us to grow more labor intensive crops? That was one of the ideas.

Josh Strine: If we are trading machinery, because of operator and the operator no longer is a constraint. Does that change used machinery values? That was, that was one of ’em.

Yeah, so it is an interesting conversation and, if you have interest or want more information about this paper. Thankfully because of Purdue’s agreement with the journal that we published it in, it is available open source. If you have interest, I would say we would both be interested in having deeper conversations if you have perspectives or things that you’re thinking about, related to sort of how is the potential of autonomy gonna impact us.

Yeah. Anytime open to have those conversations and really look forward to,being able to continue to explore this realm.

Chad Fiechter: Hopefully you’ve enjoyed this. We will put a link to the paper in the notes. Again, invite questions. And hopefully you’ll let us know if you got something you wanna talk about.